Buckets:
| <meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Text to speech","local":"text-to-speech","sections":[{"title":"Load the dataset","local":"load-the-dataset","sections":[],"depth":2},{"title":"Preprocess the data","local":"preprocess-the-data","sections":[{"title":"Text cleanup for SpeechT5 tokenization","local":"text-cleanup-for-speecht5-tokenization","sections":[],"depth":3},{"title":"Speakers","local":"speakers","sections":[],"depth":3},{"title":"Speaker embeddings","local":"speaker-embeddings","sections":[],"depth":3},{"title":"Processing the dataset","local":"processing-the-dataset","sections":[],"depth":3},{"title":"Data collator","local":"data-collator","sections":[],"depth":3}],"depth":2},{"title":"Train the model","local":"train-the-model","sections":[],"depth":2},{"title":"Inference","local":"inference","sections":[{"title":"Inference with a pipeline","local":"inference-with-a-pipeline","sections":[],"depth":3},{"title":"Run inference manually","local":"run-inference-manually","sections":[],"depth":3}],"depth":2}],"depth":1}"/> | |
| <link href="/docs/transformers/main/en/_app/immutable/entry/start.QSJpda4J.js" rel="modulepreload"> | |
| <link href="/docs/transformers/main/en/_app/immutable/chunks/CuJ5QOBg.js" rel="modulepreload"> | |
| <link href="/docs/transformers/main/en/_app/immutable/chunks/v-ay6ctG.js" rel="modulepreload"> | |
| <link href="/docs/transformers/main/en/_app/immutable/entry/app.Ch9T9LuJ.js" rel="modulepreload"> | |
| <link href="/docs/transformers/main/en/_app/immutable/chunks/BIdhRtf1.js" rel="modulepreload"> | |
| <link href="/docs/transformers/main/en/_app/immutable/chunks/Cmr6hduW.js" rel="modulepreload"> | |
| <link href="/docs/transformers/main/en/_app/immutable/chunks/B4BfNKKv.js" rel="modulepreload"> | |
| <link href="/docs/transformers/main/en/_app/immutable/chunks/DsnmJJEf.js" rel="modulepreload"> | |
| <link href="/docs/transformers/main/en/_app/immutable/nodes/0.CH7fPDr9.js" rel="modulepreload"> | |
| <link href="/docs/transformers/main/en/_app/immutable/chunks/QxajwaoF.js" rel="modulepreload"> | |
| <link href="/docs/transformers/main/en/_app/immutable/nodes/71.B8xNDdgC.js" rel="modulepreload"> | |
| <link href="/docs/transformers/main/en/_app/immutable/chunks/Qu8hRGoa.js" rel="modulepreload"> | |
| <link href="/docs/transformers/main/en/_app/immutable/chunks/Bgvn_YcA.js" rel="modulepreload"> | |
| <link href="/docs/transformers/main/en/_app/immutable/chunks/CPzph7or.js" rel="modulepreload"> | |
| <!--1k42vzo--><meta name="hf:doc:metadata" content="{"title":"Text to speech","local":"text-to-speech","sections":[{"title":"Load the dataset","local":"load-the-dataset","sections":[],"depth":2},{"title":"Preprocess the data","local":"preprocess-the-data","sections":[{"title":"Text cleanup for SpeechT5 tokenization","local":"text-cleanup-for-speecht5-tokenization","sections":[],"depth":3},{"title":"Speakers","local":"speakers","sections":[],"depth":3},{"title":"Speaker embeddings","local":"speaker-embeddings","sections":[],"depth":3},{"title":"Processing the dataset","local":"processing-the-dataset","sections":[],"depth":3},{"title":"Data collator","local":"data-collator","sections":[],"depth":3}],"depth":2},{"title":"Train the model","local":"train-the-model","sections":[],"depth":2},{"title":"Inference","local":"inference","sections":[{"title":"Inference with a pipeline","local":"inference-with-a-pipeline","sections":[],"depth":3},{"title":"Run inference manually","local":"run-inference-manually","sections":[],"depth":3}],"depth":2}],"depth":1}"/><!----> | |
| <link href="/docs/transformers/main/en/_app/immutable/assets/0.tn0RQdqM.css" rel="modulepreload"> <!--[--><!--[0--><!--[--><!--[0--><!--[--><p></p> <div class="items-center shrink-0 min-w-[100px] max-sm:min-w-[50px] justify-end ml-auto flex" style="float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"><div class="inline-flex rounded-md max-sm:rounded-sm"><button class="inline-flex items-center gap-1 h-7 max-sm:h-7 px-2 max-sm:px-1.5 text-sm font-medium text-gray-800 border border-r-0 rounded-l-md max-sm:rounded-l-sm border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-live="polite"><span class="inline-flex items-center justify-center rounded-md p-0.5 max-sm:p-0 hover:text-gray-800 dark:hover:text-gray-200"><svg class="sm:size-3.5 size-3" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----></span> <span>Copy page</span></button> <button class="inline-flex items-center justify-center w-6 max-sm:w-5 h-7 max-sm:h-7 disabled:pointer-events-none text-sm text-gray-500 hover:text-gray-700 dark:hover:text-white rounded-r-md max-sm:rounded-r-sm border border-l transition border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-haspopup="menu" aria-expanded="false" aria-label="Open copy menu"><svg class="transition-transform text-gray-400 overflow-visible sm:size-3.5 size-3 rotate-0" width="1em" height="1em" viewBox="0 0 12 7" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M1 1L6 6L11 1" stroke="currentColor"></path></svg><!----></button></div> <!--[-1--><!--]--></div><!----> <div class="flex space-x-1 " style="float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"><!----> <!--[1--><div class="relative colab-dropdown "><button class=" " type="button"><!--[0--><!--[-1--><img alt="Open In Colab" class="!m-0" src="https://colab.research.google.com/assets/colab-badge.svg"/><!--]--><!----><!--]--></button> <!--[-1--><!--]--></div><!--]--> <!--[1--><div class="relative colab-dropdown "><button class=" " type="button"><!--[0--><!--[-1--><img alt="Open In Studio Lab" class="!m-0" src="https://studiolab.sagemaker.aws/studiolab.svg"/><!--]--><!----><!--]--></button> <!--[-1--><!--]--></div><!--]--></div><!----> <!--[0--><h1 class="relative group"><a id="text-to-speech" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#text-to-speech"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span>Text to speech</span></h1><!--]--><!----> <p>Text-to-speech (TTS) is the task of creating natural-sounding speech from text, where the speech can be generated in multiple | |
| languages and for multiple speakers. Several text-to-speech models are currently available in 🤗 Transformers, such as <a href="../model_doc/dia">Dia</a>, <a href="../model_doc/csm">CSM</a>, <a href="../model_doc/bark">Bark</a>, <a href="../model_doc/mms">MMS</a>, <a href="../model_doc/vits">VITS</a> and <a href="../model_doc/speecht5">SpeechT5</a>.</p> <p>You can easily generate audio using the <code>"text-to-audio"</code> pipeline (or its alias - <code>"text-to-speech"</code>). | |
| Here’s an example of how you would use the <code>"text-to-speech"</code> pipeline with <a href="https://huggingface.co/sesame/csm-1b" rel="nofollow">CSM</a>:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> pipeline | |
| <span class="hljs-meta">>>> </span>pipe = pipeline(<span class="hljs-string">"text-to-audio"</span>, model=<span class="hljs-string">"sesame/csm-1b"</span>) | |
| <span class="hljs-meta">>>> </span>output = pipe(<span class="hljs-string">"Hello from Sesame."</span>)<!----></pre></div><!----> <p>Here’s a code snippet you can use to listen to the resulting audio in a notebook:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> IPython.display <span class="hljs-keyword">import</span> Audio | |
| <span class="hljs-meta">>>> </span>Audio(output[<span class="hljs-string">"audio"</span>], rate=output[<span class="hljs-string">"sampling_rate"</span>])<!----></pre></div><!----> <p>By default, CSM uses a random voice. You can do voice cloning by providing a reference audio as part of a chat template dictionary:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> soundfile <span class="hljs-keyword">as</span> sf | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> datasets <span class="hljs-keyword">import</span> Audio, load_dataset | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> pipeline | |
| <span class="hljs-meta">>>> </span>pipe = pipeline(<span class="hljs-string">"text-to-audio"</span>, model=<span class="hljs-string">"sesame/csm-1b"</span>) | |
| <span class="hljs-meta">>>> </span>ds = load_dataset(<span class="hljs-string">"hf-internal-testing/dailytalk-dummy"</span>, split=<span class="hljs-string">"train"</span>) | |
| <span class="hljs-meta">>>> </span>ds = ds.cast_column(<span class="hljs-string">"audio"</span>, Audio(sampling_rate=<span class="hljs-number">24000</span>)) | |
| <span class="hljs-meta">>>> </span>conversation = [ | |
| <span class="hljs-meta">... </span> { | |
| <span class="hljs-meta">... </span> <span class="hljs-string">"role"</span>: <span class="hljs-string">"0"</span>, | |
| <span class="hljs-meta">... </span> <span class="hljs-string">"content"</span>: [ | |
| <span class="hljs-meta">... </span> {<span class="hljs-string">"type"</span>: <span class="hljs-string">"text"</span>, <span class="hljs-string">"text"</span>: <span class="hljs-string">"What are you working on?"</span>}, | |
| <span class="hljs-meta">... </span> {<span class="hljs-string">"type"</span>: <span class="hljs-string">"audio"</span>, <span class="hljs-string">"path"</span>: ds[<span class="hljs-number">0</span>][<span class="hljs-string">"audio"</span>][<span class="hljs-string">"array"</span>]}, | |
| <span class="hljs-meta">... </span> ], | |
| <span class="hljs-meta">... </span> }, | |
| <span class="hljs-meta">... </span> {<span class="hljs-string">"role"</span>: <span class="hljs-string">"0"</span>, <span class="hljs-string">"content"</span>: [{<span class="hljs-string">"type"</span>: <span class="hljs-string">"text"</span>, <span class="hljs-string">"text"</span>: <span class="hljs-string">"How much money can you spend?"</span>}]}, | |
| <span class="hljs-meta">... </span>] | |
| <span class="hljs-meta">>>> </span>output = pipe(conversation)<!----></pre></div><!----> <p>Some models, like <a href="https://huggingface.co/nari-labs/Dia-1.6B-0626" rel="nofollow">Dia</a>, can also be conditioned to generate non-verbal communications such as laughing, sighing and crying, or even add music. Below is such an example:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> pipeline | |
| <span class="hljs-meta">>>> </span>pipe = pipeline(<span class="hljs-string">"text-to-speech"</span>, model=<span class="hljs-string">"nari-labs/Dia-1.6B-0626"</span>) | |
| <span class="hljs-meta">>>> </span>text = <span class="hljs-string">"[S1] (clears throat) Hello! How are you? [S2] I'm good, thanks! How about you?"</span> | |
| <span class="hljs-meta">>>> </span>output = pipe(text)<!----></pre></div><!----> <p>Note that Dia also accepts speaker tags such as [S1] and [S2] to generate a conversation between unique voices.</p> <p>For more examples on what CSM and other pretrained TTS models can do, refer to our <a href="https://huggingface.co/learn/audio-course/chapter6/pre-trained_models" rel="nofollow">Audio course</a>.</p> <p>If you are looking to fine-tune a TTS model, the only text-to-speech models currently available in 🤗 Transformers | |
| are <a href="../model_doc/speecht5">SpeechT5</a>, <a href="../model_doc/fastspeech2_conformer">FastSpeech2Conformer</a>, <a href="../model_doc/dia">Dia</a> and <a href="../model_doc/csm">CSM</a> though more will be added in the future. SpeechT5 is pre-trained on a combination of speech-to-text and text-to-speech data, allowing it to learn a unified space of hidden representations shared by both text and speech. This means that the same pre-trained model can be fine-tuned for different tasks. Furthermore, SpeechT5 supports multiple speakers through x-vector speaker embeddings.</p> <p>The remainder of this guide illustrates how to:</p> <ol><li>Fine-tune <a href="../model_doc/speecht5">SpeechT5</a> that was originally trained on English speech on the Dutch (<code>nl</code>) language subset of the <a href="https://huggingface.co/datasets/qmeeus/voxpopuli" rel="nofollow">VoxPopuli</a> dataset.</li> <li>Use your refined model for inference in one of two ways: using a pipeline or directly.</li></ol> <p>Before you begin, make sure you have all the necessary libraries installed:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-bash "><!---->pip install transformers datasets soundfile librosa torchcodec speechbrain accelerate<!----></pre></div><!----> <p>Install 🤗Transformers from source as not all the SpeechT5 features have been merged into an official release yet:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-bash "><!---->pip install git+https://github.com/huggingface/transformers.git<!----></pre></div><!----> <blockquote class="tip"><p>To follow this guide you will need a GPU. If you’re working in a notebook, run the following line to check if a GPU is available:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-bash "><!---->!nvidia-smi<!----></pre></div><!----> <p>or alternatively for AMD GPUs:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-bash "><!---->!rocm-smi<!----></pre></div><!----><!----></blockquote><!----> <p>We encourage you to log in to your Hugging Face account to upload and share your model with the community. When prompted, enter your token to log in:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> huggingface_hub <span class="hljs-keyword">import</span> notebook_login | |
| <span class="hljs-meta">>>> </span>notebook_login()<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="load-the-dataset" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#load-the-dataset"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span>Load the dataset</span></h2><!--]--><!----> <p><a href="https://huggingface.co/datasets/qmeeus/voxpopuli" rel="nofollow">VoxPopuli</a> is a large-scale multilingual speech corpus consisting of | |
| data sourced from 2009-2020 European Parliament event recordings. It contains labelled audio-transcription data for 15 | |
| European languages. In this guide, we are using the Dutch language subset, feel free to pick another subset.</p> <p>Note that VoxPopuli or any other automated speech recognition (ASR) dataset may not be the most suitable | |
| option for training TTS models. The features that make it beneficial for ASR, such as excessive background noise, are | |
| typically undesirable in TTS. However, finding top-quality, multilingual, and multi-speaker TTS datasets can be quite | |
| challenging.</p> <p>Let’s load the data:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> datasets <span class="hljs-keyword">import</span> load_dataset, Audio | |
| <span class="hljs-meta">>>> </span>dataset = load_dataset(<span class="hljs-string">"qmeeus/voxpopuli"</span>, <span class="hljs-string">"nl"</span>, split=<span class="hljs-string">"train"</span>) | |
| <span class="hljs-meta">>>> </span><span class="hljs-built_in">len</span>(dataset) | |
| <span class="hljs-number">20968</span><!----></pre></div><!----> <p>20968 examples should be sufficient for fine-tuning. SpeechT5 expects audio data to have a sampling rate of 16 kHz, so | |
| make sure the examples in the dataset meet this requirement:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!---->dataset = dataset.cast_column(<span class="hljs-string">"audio"</span>, Audio(sampling_rate=<span class="hljs-number">16000</span>))<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="preprocess-the-data" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#preprocess-the-data"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span>Preprocess the data</span></h2><!--]--><!----> <p>Let’s begin by defining the model checkpoint to use and loading the appropriate processor:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> SpeechT5Processor | |
| <span class="hljs-meta">>>> </span>checkpoint = <span class="hljs-string">"microsoft/speecht5_tts"</span> | |
| <span class="hljs-meta">>>> </span>processor = SpeechT5Processor.from_pretrained(checkpoint)<!----></pre></div><!----> <!--[2--><h3 class="relative group"><a id="text-cleanup-for-speecht5-tokenization" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#text-cleanup-for-speecht5-tokenization"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span>Text cleanup for SpeechT5 tokenization</span></h3><!--]--><!----> <p>Start by cleaning up the text data. You’ll need the tokenizer part of the processor to process the text:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>tokenizer = processor.tokenizer<!----></pre></div><!----> <p>The dataset examples contain a <code>text</code> feature with the transcription. Note that the SpeechT5 tokenizer doesn’t have any tokens for numbers, so any numbers in the text will be converted to <code><unk></code> tokens. If your dataset contains numbers, consider normalizing them to their written form.</p> <p>Because SpeechT5 was trained on the English language, it may not recognize certain characters in the Dutch dataset. If | |
| left as is, these characters will be converted to <code><unk></code> tokens. However, in Dutch, certain characters like <code>à</code> are | |
| used to stress syllables. In order to preserve the meaning of the text, we can replace this character with a regular <code>a</code>.</p> <p>To identify unsupported tokens, extract all unique characters in the dataset using the <code>SpeechT5Tokenizer</code> which | |
| works with characters as tokens. To do this, write the <code>extract_all_chars</code> mapping function that concatenates | |
| the transcriptions from all examples into one string and converts it to a set of characters. | |
| Make sure to set <code>batched=True</code> and <code>batch_size=-1</code> in <code>dataset.map()</code> so that all transcriptions are available at once for | |
| the mapping function.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">def</span> <span class="hljs-title function_">extract_all_chars</span>(<span class="hljs-params">batch</span>): | |
| <span class="hljs-meta">... </span> all_text = <span class="hljs-string">" "</span>.join(batch[<span class="hljs-string">"text"</span>]) | |
| <span class="hljs-meta">... </span> vocab = <span class="hljs-built_in">list</span>(<span class="hljs-built_in">set</span>(all_text)) | |
| <span class="hljs-meta">... </span> <span class="hljs-keyword">return</span> {<span class="hljs-string">"vocab"</span>: [vocab], <span class="hljs-string">"all_text"</span>: [all_text]} | |
| <span class="hljs-meta">>>> </span>vocabs = dataset.<span class="hljs-built_in">map</span>( | |
| <span class="hljs-meta">... </span> extract_all_chars, | |
| <span class="hljs-meta">... </span> batched=<span class="hljs-literal">True</span>, | |
| <span class="hljs-meta">... </span> batch_size=-<span class="hljs-number">1</span>, | |
| <span class="hljs-meta">... </span> keep_in_memory=<span class="hljs-literal">True</span>, | |
| <span class="hljs-meta">... </span> remove_columns=dataset.column_names, | |
| <span class="hljs-meta">... </span>) | |
| <span class="hljs-meta">>>> </span>dataset_vocab = <span class="hljs-built_in">set</span>(vocabs[<span class="hljs-string">"vocab"</span>][<span class="hljs-number">0</span>]) | |
| <span class="hljs-meta">>>> </span>tokenizer_vocab = {k <span class="hljs-keyword">for</span> k, _ <span class="hljs-keyword">in</span> tokenizer.get_vocab().items()}<!----></pre></div><!----> <p>Now you have two sets of characters: one with the vocabulary from the dataset and one with the vocabulary from the tokenizer. | |
| To identify any unsupported characters in the dataset, you can take the difference between these two sets. The resulting | |
| set will contain the characters that are in the dataset but not in the tokenizer.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>dataset_vocab - tokenizer_vocab | |
| {<span class="hljs-string">' '</span>, <span class="hljs-string">'à'</span>, <span class="hljs-string">'ç'</span>, <span class="hljs-string">'è'</span>, <span class="hljs-string">'ë'</span>, <span class="hljs-string">'í'</span>, <span class="hljs-string">'ï'</span>, <span class="hljs-string">'ö'</span>, <span class="hljs-string">'ü'</span>}<!----></pre></div><!----> <p>To handle the unsupported characters identified in the previous step, define a function that maps these characters to | |
| valid tokens. Note that spaces are already replaced by <code>▁</code> in the tokenizer and don’t need to be handled separately.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>replacements = [ | |
| <span class="hljs-meta">... </span> (<span class="hljs-string">"à"</span>, <span class="hljs-string">"a"</span>), | |
| <span class="hljs-meta">... </span> (<span class="hljs-string">"ç"</span>, <span class="hljs-string">"c"</span>), | |
| <span class="hljs-meta">... </span> (<span class="hljs-string">"è"</span>, <span class="hljs-string">"e"</span>), | |
| <span class="hljs-meta">... </span> (<span class="hljs-string">"ë"</span>, <span class="hljs-string">"e"</span>), | |
| <span class="hljs-meta">... </span> (<span class="hljs-string">"í"</span>, <span class="hljs-string">"i"</span>), | |
| <span class="hljs-meta">... </span> (<span class="hljs-string">"ï"</span>, <span class="hljs-string">"i"</span>), | |
| <span class="hljs-meta">... </span> (<span class="hljs-string">"ö"</span>, <span class="hljs-string">"o"</span>), | |
| <span class="hljs-meta">... </span> (<span class="hljs-string">"ü"</span>, <span class="hljs-string">"u"</span>), | |
| <span class="hljs-meta">... </span>] | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">def</span> <span class="hljs-title function_">cleanup_text</span>(<span class="hljs-params">inputs</span>): | |
| <span class="hljs-meta">... </span> <span class="hljs-keyword">for</span> src, dst <span class="hljs-keyword">in</span> replacements: | |
| <span class="hljs-meta">... </span> inputs[<span class="hljs-string">"text"</span>] = inputs[<span class="hljs-string">"text"</span>].replace(src, dst) | |
| <span class="hljs-meta">... </span> <span class="hljs-keyword">return</span> inputs | |
| <span class="hljs-meta">>>> </span>dataset = dataset.<span class="hljs-built_in">map</span>(cleanup_text)<!----></pre></div><!----> <p>Now that you have dealt with special characters in the text, it’s time to shift focus to the audio data.</p> <!--[2--><h3 class="relative group"><a id="speakers" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#speakers"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span>Speakers</span></h3><!--]--><!----> <p>The VoxPopuli dataset includes speech from multiple speakers, but how many speakers are represented in the dataset? To | |
| determine this, we can count the number of unique speakers and the number of examples each speaker contributes to the dataset. | |
| With a total of 20,968 examples in the dataset, this information will give us a better understanding of the distribution of | |
| speakers and examples in the data.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> collections <span class="hljs-keyword">import</span> defaultdict | |
| <span class="hljs-meta">>>> </span>speaker_counts = defaultdict(<span class="hljs-built_in">int</span>) | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">for</span> speaker_id <span class="hljs-keyword">in</span> dataset[<span class="hljs-string">"speaker_id"</span>]: | |
| <span class="hljs-meta">... </span> speaker_counts[speaker_id] += <span class="hljs-number">1</span><!----></pre></div><!----> <p>By plotting a histogram you can get a sense of how much data there is for each speaker.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> matplotlib.pyplot <span class="hljs-keyword">as</span> plt | |
| <span class="hljs-meta">>>> </span>plt.figure() | |
| <span class="hljs-meta">>>> </span>plt.hist(speaker_counts.values(), bins=<span class="hljs-number">20</span>) | |
| <span class="hljs-meta">>>> </span>plt.ylabel(<span class="hljs-string">"Speakers"</span>) | |
| <span class="hljs-meta">>>> </span>plt.xlabel(<span class="hljs-string">"Examples"</span>) | |
| <span class="hljs-meta">>>> </span>plt.show()<!----></pre></div><!----> <div class="flex justify-center"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/tts_speakers_histogram.png" alt="Speakers histogram"/></div> <p>The histogram reveals that approximately one-third of the speakers in the dataset have fewer than 100 examples, while | |
| around ten speakers have more than 500 examples. To improve training efficiency and balance the dataset, we can limit | |
| the data to speakers with between 100 and 400 examples.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">def</span> <span class="hljs-title function_">select_speaker</span>(<span class="hljs-params">speaker_id</span>): | |
| <span class="hljs-meta">... </span> <span class="hljs-keyword">return</span> <span class="hljs-number">100</span> <= speaker_counts[speaker_id] <= <span class="hljs-number">400</span> | |
| <span class="hljs-meta">>>> </span>dataset = dataset.<span class="hljs-built_in">filter</span>(select_speaker, input_columns=[<span class="hljs-string">"speaker_id"</span>])<!----></pre></div><!----> <p>Let’s check how many speakers remain:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-built_in">len</span>(<span class="hljs-built_in">set</span>(dataset[<span class="hljs-string">"speaker_id"</span>])) | |
| <span class="hljs-number">42</span><!----></pre></div><!----> <p>Let’s see how many examples are left:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-built_in">len</span>(dataset) | |
| <span class="hljs-number">9973</span><!----></pre></div><!----> <p>You are left with just under 10,000 examples from approximately 40 unique speakers, which should be sufficient.</p> <p>Note that some speakers with few examples may actually have more audio available if the examples are long. However, | |
| determining the total amount of audio for each speaker requires scanning through the entire dataset, which is a | |
| time-consuming process that involves loading and decoding each audio file. As such, we have chosen to skip this step here.</p> <!--[2--><h3 class="relative group"><a id="speaker-embeddings" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#speaker-embeddings"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span>Speaker embeddings</span></h3><!--]--><!----> <p>To enable the TTS model to differentiate between multiple speakers, you’ll need to create a speaker embedding for each example. | |
| The speaker embedding is an additional input into the model that captures a particular speaker’s voice characteristics. | |
| To generate these speaker embeddings, use the pre-trained <a href="https://huggingface.co/speechbrain/spkrec-xvect-voxceleb" rel="nofollow">spkrec-xvect-voxceleb</a> model from SpeechBrain.</p> <p>Create a function <code>create_speaker_embedding()</code> that takes an input audio waveform and outputs a 512-element vector | |
| containing the corresponding speaker embedding.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> os | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> torch | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> speechbrain.inference.classifiers <span class="hljs-keyword">import</span> EncoderClassifier | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> accelerate <span class="hljs-keyword">import</span> Accelerator | |
| <span class="hljs-meta">>>> </span>spk_model_name = <span class="hljs-string">"speechbrain/spkrec-xvect-voxceleb"</span> | |
| <span class="hljs-meta">>>> </span>device = Accelerator().device | |
| <span class="hljs-meta">>>> </span>speaker_model = EncoderClassifier.from_hparams( | |
| <span class="hljs-meta">... </span> source=spk_model_name, | |
| <span class="hljs-meta">... </span> run_opts={<span class="hljs-string">"device"</span>: device}, | |
| <span class="hljs-meta">... </span> savedir=os.path.join(<span class="hljs-string">"/tmp"</span>, spk_model_name), | |
| <span class="hljs-meta">... </span>) | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">def</span> <span class="hljs-title function_">create_speaker_embedding</span>(<span class="hljs-params">waveform</span>): | |
| <span class="hljs-meta">... </span> <span class="hljs-keyword">with</span> torch.no_grad(): | |
| <span class="hljs-meta">... </span> speaker_embeddings = speaker_model.encode_batch(torch.tensor(waveform)) | |
| <span class="hljs-meta">... </span> speaker_embeddings = torch.nn.functional.normalize(speaker_embeddings, dim=<span class="hljs-number">2</span>) | |
| <span class="hljs-meta">... </span> speaker_embeddings = speaker_embeddings.squeeze().cpu().numpy() | |
| <span class="hljs-meta">... </span> <span class="hljs-keyword">return</span> speaker_embeddings<!----></pre></div><!----> <p>It’s important to note that the <code>speechbrain/spkrec-xvect-voxceleb</code> model was trained on English speech from the VoxCeleb | |
| dataset, whereas the training examples in this guide are in Dutch. While we believe that this model will still generate | |
| reasonable speaker embeddings for our Dutch dataset, this assumption may not hold true in all cases.</p> <p>For optimal results, we recommend training an X-vector model on the target speech first. This will ensure that the model | |
| is better able to capture the unique voice characteristics present in the Dutch language.</p> <!--[2--><h3 class="relative group"><a id="processing-the-dataset" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#processing-the-dataset"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span>Processing the dataset</span></h3><!--]--><!----> <p>Finally, let’s process the data into the format the model expects. Create a <code>prepare_dataset</code> function that takes in a | |
| single example and uses the <code>SpeechT5Processor</code> object to tokenize the input text and load the target audio into a log-mel spectrogram. | |
| It should also add the speaker embeddings as an additional input.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">def</span> <span class="hljs-title function_">prepare_dataset</span>(<span class="hljs-params">example</span>): | |
| <span class="hljs-meta">... </span> audio = example[<span class="hljs-string">"audio"</span>] | |
| <span class="hljs-meta">... </span> example = processor( | |
| <span class="hljs-meta">... </span> text=example[<span class="hljs-string">"text"</span>], | |
| <span class="hljs-meta">... </span> audio_target=audio[<span class="hljs-string">"array"</span>], | |
| <span class="hljs-meta">... </span> sampling_rate=audio[<span class="hljs-string">"sampling_rate"</span>], | |
| <span class="hljs-meta">... </span> return_attention_mask=<span class="hljs-literal">False</span>, | |
| <span class="hljs-meta">... </span> ) | |
| <span class="hljs-meta">... </span> <span class="hljs-comment"># strip off the batch dimension</span> | |
| <span class="hljs-meta">... </span> example[<span class="hljs-string">"labels"</span>] = example[<span class="hljs-string">"labels"</span>][<span class="hljs-number">0</span>] | |
| <span class="hljs-meta">... </span> <span class="hljs-comment"># use SpeechBrain to obtain x-vector</span> | |
| <span class="hljs-meta">... </span> example[<span class="hljs-string">"speaker_embeddings"</span>] = create_speaker_embedding(audio[<span class="hljs-string">"array"</span>]) | |
| <span class="hljs-meta">... </span> <span class="hljs-keyword">return</span> example<!----></pre></div><!----> <p>Verify the processing is correct by looking at a single example:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>processed_example = prepare_dataset(dataset[<span class="hljs-number">0</span>]) | |
| <span class="hljs-meta">>>> </span><span class="hljs-built_in">list</span>(processed_example.keys()) | |
| [<span class="hljs-string">'input_ids'</span>, <span class="hljs-string">'labels'</span>, <span class="hljs-string">'stop_labels'</span>, <span class="hljs-string">'speaker_embeddings'</span>]<!----></pre></div><!----> <p>Speaker embeddings should be a 512-element vector:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>processed_example[<span class="hljs-string">"speaker_embeddings"</span>].shape | |
| (<span class="hljs-number">512</span>,)<!----></pre></div><!----> <p>The labels should be a log-mel spectrogram with 80 mel bins.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">import</span> matplotlib.pyplot <span class="hljs-keyword">as</span> plt | |
| <span class="hljs-meta">>>> </span>plt.figure() | |
| <span class="hljs-meta">>>> </span>plt.imshow(processed_example[<span class="hljs-string">"labels"</span>].T) | |
| <span class="hljs-meta">>>> </span>plt.show()<!----></pre></div><!----> <div class="flex justify-center"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/tts_logmelspectrogram_1.png" alt="Log-mel spectrogram with 80 mel bins"/></div> <p>Side note: If you find this spectrogram confusing, it may be due to your familiarity with the convention of placing low frequencies | |
| at the bottom and high frequencies at the top of a plot. However, when plotting spectrograms as an image using the matplotlib library, | |
| the y-axis is flipped and the spectrograms appear upside down.</p> <p>Now apply the processing function to the entire dataset. This will take between 5 and 10 minutes.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>dataset = dataset.<span class="hljs-built_in">map</span>(prepare_dataset, remove_columns=dataset.column_names)<!----></pre></div><!----> <p>You’ll see a warning saying that some examples in the dataset are longer than the maximum input length the model can handle (600 tokens). | |
| Remove those examples from the dataset. Here we go even further and to allow for larger batch sizes we remove anything over 200 tokens.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">def</span> <span class="hljs-title function_">is_not_too_long</span>(<span class="hljs-params">input_ids</span>): | |
| <span class="hljs-meta">... </span> input_length = <span class="hljs-built_in">len</span>(input_ids) | |
| <span class="hljs-meta">... </span> <span class="hljs-keyword">return</span> input_length < <span class="hljs-number">200</span> | |
| <span class="hljs-meta">>>> </span>dataset = dataset.<span class="hljs-built_in">filter</span>(is_not_too_long, input_columns=[<span class="hljs-string">"input_ids"</span>]) | |
| <span class="hljs-meta">>>> </span><span class="hljs-built_in">len</span>(dataset) | |
| <span class="hljs-number">8259</span><!----></pre></div><!----> <p>Next, create a basic train/test split:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>dataset = dataset.train_test_split(test_size=<span class="hljs-number">0.1</span>)<!----></pre></div><!----> <!--[2--><h3 class="relative group"><a id="data-collator" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#data-collator"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span>Data collator</span></h3><!--]--><!----> <p>In order to combine multiple examples into a batch, you need to define a custom data collator. This collator will pad shorter sequences with padding | |
| tokens, ensuring that all examples have the same length. For the spectrogram labels, the padded portions are replaced with the special value <code>-100</code>. This special value | |
| instructs the model to ignore that part of the spectrogram when calculating the spectrogram loss.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> dataclasses <span class="hljs-keyword">import</span> dataclass | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> typing <span class="hljs-keyword">import</span> <span class="hljs-type">Any</span>, <span class="hljs-type">Dict</span>, <span class="hljs-type">List</span>, <span class="hljs-type">Union</span> | |
| <span class="hljs-meta">>>> </span>@dataclass | |
| <span class="hljs-meta">... </span><span class="hljs-keyword">class</span> <span class="hljs-title class_">TTSDataCollatorWithPadding</span>: | |
| <span class="hljs-meta">... </span> processor: <span class="hljs-type">Any</span> | |
| <span class="hljs-meta">... </span> <span class="hljs-keyword">def</span> <span class="hljs-title function_">__call__</span>(<span class="hljs-params">self, features: <span class="hljs-built_in">list</span>[<span class="hljs-built_in">dict</span>[<span class="hljs-built_in">str</span>, <span class="hljs-type">Union</span>[<span class="hljs-built_in">list</span>[<span class="hljs-built_in">int</span>], torch.Tensor]]]</span>) -> <span class="hljs-built_in">dict</span>[<span class="hljs-built_in">str</span>, torch.Tensor]: | |
| <span class="hljs-meta">... </span> input_ids = [{<span class="hljs-string">"input_ids"</span>: feature[<span class="hljs-string">"input_ids"</span>]} <span class="hljs-keyword">for</span> feature <span class="hljs-keyword">in</span> features] | |
| <span class="hljs-meta">... </span> label_features = [{<span class="hljs-string">"input_values"</span>: feature[<span class="hljs-string">"labels"</span>]} <span class="hljs-keyword">for</span> feature <span class="hljs-keyword">in</span> features] | |
| <span class="hljs-meta">... </span> speaker_features = [feature[<span class="hljs-string">"speaker_embeddings"</span>] <span class="hljs-keyword">for</span> feature <span class="hljs-keyword">in</span> features] | |
| <span class="hljs-meta">... </span> <span class="hljs-comment"># collate the inputs and targets into a batch</span> | |
| <span class="hljs-meta">... </span> batch = processor.pad(input_ids=input_ids, labels=label_features, return_tensors=<span class="hljs-string">"pt"</span>) | |
| <span class="hljs-meta">... </span> <span class="hljs-comment"># replace padding with -100 to ignore loss correctly</span> | |
| <span class="hljs-meta">... </span> batch[<span class="hljs-string">"labels"</span>] = batch[<span class="hljs-string">"labels"</span>].masked_fill(batch.decoder_attention_mask.unsqueeze(-<span class="hljs-number">1</span>).ne(<span class="hljs-number">1</span>), -<span class="hljs-number">100</span>) | |
| <span class="hljs-meta">... </span> <span class="hljs-comment"># not used during fine-tuning</span> | |
| <span class="hljs-meta">... </span> <span class="hljs-keyword">del</span> batch[<span class="hljs-string">"decoder_attention_mask"</span>] | |
| <span class="hljs-meta">... </span> <span class="hljs-comment"># round down target lengths to multiple of reduction factor</span> | |
| <span class="hljs-meta">... </span> <span class="hljs-keyword">if</span> model.config.reduction_factor > <span class="hljs-number">1</span>: | |
| <span class="hljs-meta">... </span> target_lengths = torch.tensor([<span class="hljs-built_in">len</span>(feature[<span class="hljs-string">"input_values"</span>]) <span class="hljs-keyword">for</span> feature <span class="hljs-keyword">in</span> label_features]) | |
| <span class="hljs-meta">... </span> target_lengths = target_lengths.new( | |
| <span class="hljs-meta">... </span> [length - length % model.config.reduction_factor <span class="hljs-keyword">for</span> length <span class="hljs-keyword">in</span> target_lengths] | |
| <span class="hljs-meta">... </span> ) | |
| <span class="hljs-meta">... </span> max_length = <span class="hljs-built_in">max</span>(target_lengths) | |
| <span class="hljs-meta">... </span> batch[<span class="hljs-string">"labels"</span>] = batch[<span class="hljs-string">"labels"</span>][:, :max_length] | |
| <span class="hljs-meta">... </span> <span class="hljs-comment"># also add in the speaker embeddings</span> | |
| <span class="hljs-meta">... </span> batch[<span class="hljs-string">"speaker_embeddings"</span>] = torch.tensor(speaker_features) | |
| <span class="hljs-meta">... </span> <span class="hljs-keyword">return</span> batch<!----></pre></div><!----> <p>In SpeechT5, the input to the decoder part of the model is reduced by a factor 2. In other words, it throws away every | |
| other timestep from the target sequence. The decoder then predicts a sequence that is twice as long. Since the original | |
| target sequence length may be odd, the data collator makes sure to round the maximum length of the batch down to be a | |
| multiple of 2.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>data_collator = TTSDataCollatorWithPadding(processor=processor)<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="train-the-model" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#train-the-model"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span>Train the model</span></h2><!--]--><!----> <p>Load the pre-trained model from the same checkpoint as you used for loading the processor:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> SpeechT5ForTextToSpeech | |
| <span class="hljs-meta">>>> </span>model = SpeechT5ForTextToSpeech.from_pretrained(checkpoint)<!----></pre></div><!----> <p>The <code>use_cache=True</code> option is incompatible with gradient checkpointing. Disable it for training.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>model.config.use_cache = <span class="hljs-literal">False</span><!----></pre></div><!----> <p>Define the training arguments. Here we are not computing any evaluation metrics during the training process. Instead, we’ll | |
| only look at the loss:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> Seq2SeqTrainingArguments | |
| <span class="hljs-meta">>>> </span>training_args = Seq2SeqTrainingArguments( | |
| <span class="hljs-meta">... </span> output_dir=<span class="hljs-string">"speecht5_finetuned_voxpopuli_nl"</span>, <span class="hljs-comment"># change to a repo name of your choice</span> | |
| <span class="hljs-meta">... </span> per_device_train_batch_size=<span class="hljs-number">4</span>, | |
| <span class="hljs-meta">... </span> gradient_accumulation_steps=<span class="hljs-number">8</span>, | |
| <span class="hljs-meta">... </span> learning_rate=<span class="hljs-number">1e-5</span>, | |
| <span class="hljs-meta">... </span> warmup_steps=<span class="hljs-number">500</span>, | |
| <span class="hljs-meta">... </span> max_steps=<span class="hljs-number">4000</span>, | |
| <span class="hljs-meta">... </span> gradient_checkpointing=<span class="hljs-literal">True</span>, | |
| <span class="hljs-meta">... </span> fp16=<span class="hljs-literal">True</span>, | |
| <span class="hljs-meta">... </span> eval_strategy=<span class="hljs-string">"steps"</span>, | |
| <span class="hljs-meta">... </span> per_device_eval_batch_size=<span class="hljs-number">2</span>, | |
| <span class="hljs-meta">... </span> save_steps=<span class="hljs-number">1000</span>, | |
| <span class="hljs-meta">... </span> eval_steps=<span class="hljs-number">1000</span>, | |
| <span class="hljs-meta">... </span> logging_steps=<span class="hljs-number">25</span>, | |
| <span class="hljs-meta">... </span> report_to=<span class="hljs-string">"trackio"</span>, | |
| <span class="hljs-meta">... </span> load_best_model_at_end=<span class="hljs-literal">True</span>, | |
| <span class="hljs-meta">... </span> greater_is_better=<span class="hljs-literal">False</span>, | |
| <span class="hljs-meta">... </span> label_names=[<span class="hljs-string">"labels"</span>], | |
| <span class="hljs-meta">... </span> push_to_hub=<span class="hljs-literal">True</span>, | |
| <span class="hljs-meta">... </span>)<!----></pre></div><!----> <p>Instantiate the <code>Trainer</code> object and pass the model, dataset, and data collator to it.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> Seq2SeqTrainer | |
| <span class="hljs-meta">>>> </span>trainer = Seq2SeqTrainer( | |
| <span class="hljs-meta">... </span> args=training_args, | |
| <span class="hljs-meta">... </span> model=model, | |
| <span class="hljs-meta">... </span> train_dataset=dataset[<span class="hljs-string">"train"</span>], | |
| <span class="hljs-meta">... </span> eval_dataset=dataset[<span class="hljs-string">"test"</span>], | |
| <span class="hljs-meta">... </span> data_collator=data_collator, | |
| <span class="hljs-meta">... </span> processing_class=processor, | |
| <span class="hljs-meta">... </span>)<!----></pre></div><!----> <p>And with that, you’re ready to start training! Training will take several hours. Depending on your GPU, | |
| it is possible that you will encounter a CUDA “out-of-memory” error when you start training. In this case, you can reduce | |
| the <code>per_device_train_batch_size</code> incrementally by factors of 2 and increase <code>gradient_accumulation_steps</code> by 2x to compensate.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>trainer.train()<!----></pre></div><!----> <p>To be able to use your checkpoint with a pipeline, make sure to save the processor with the checkpoint:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>processor.save_pretrained(<span class="hljs-string">"YOUR_ACCOUNT_NAME/speecht5_finetuned_voxpopuli_nl"</span>)<!----></pre></div><!----> <p>Push the final model to the 🤗 Hub:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>trainer.push_to_hub()<!----></pre></div><!----> <!--[1--><h2 class="relative group"><a id="inference" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#inference"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span>Inference</span></h2><!--]--><!----> <!--[2--><h3 class="relative group"><a id="inference-with-a-pipeline" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#inference-with-a-pipeline"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span>Inference with a pipeline</span></h3><!--]--><!----> <p>Great, now that you’ve fine-tuned a model, you can use it for inference! | |
| First, let’s see how you can use it with a corresponding pipeline. Let’s create a <code>"text-to-speech"</code> pipeline with your | |
| checkpoint:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> pipeline | |
| <span class="hljs-meta">>>> </span>pipe = pipeline(<span class="hljs-string">"text-to-speech"</span>, model=<span class="hljs-string">"YOUR_ACCOUNT_NAME/speecht5_finetuned_voxpopuli_nl"</span>)<!----></pre></div><!----> <p>Pick a piece of text in Dutch you’d like narrated, e.g.:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>text = <span class="hljs-string">"hallo allemaal, ik praat nederlands. groetjes aan iedereen!"</span><!----></pre></div><!----> <p>To use SpeechT5 with the pipeline, you’ll need a speaker embedding. Let’s get it from an example in the test dataset:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>example = dataset[<span class="hljs-string">"test"</span>][<span class="hljs-number">304</span>] | |
| <span class="hljs-meta">>>> </span>speaker_embeddings = torch.tensor(example[<span class="hljs-string">"speaker_embeddings"</span>]).unsqueeze(<span class="hljs-number">0</span>)<!----></pre></div><!----> <p>Now you can pass the text and speaker embeddings to the pipeline, and it will take care of the rest:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>forward_params = {<span class="hljs-string">"speaker_embeddings"</span>: speaker_embeddings} | |
| <span class="hljs-meta">>>> </span>output = pipe(text, forward_params=forward_params) | |
| <span class="hljs-meta">>>> </span>output | |
| {<span class="hljs-string">'audio'</span>: array([-<span class="hljs-number">6.82714235e-05</span>, -<span class="hljs-number">4.26525949e-04</span>, <span class="hljs-number">1.06134125e-04</span>, ..., | |
| -<span class="hljs-number">1.22392643e-03</span>, -<span class="hljs-number">7.76011671e-04</span>, <span class="hljs-number">3.29112721e-04</span>], dtype=float32), | |
| <span class="hljs-string">'sampling_rate'</span>: <span class="hljs-number">16000</span>}<!----></pre></div><!----> <p>You can then listen to the result:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> IPython.display <span class="hljs-keyword">import</span> Audio | |
| <span class="hljs-meta">>>> </span>Audio(output[<span class="hljs-string">'audio'</span>], rate=output[<span class="hljs-string">'sampling_rate'</span>])<!----></pre></div><!----> <!--[2--><h3 class="relative group"><a id="run-inference-manually" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#run-inference-manually"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span>Run inference manually</span></h3><!--]--><!----> <p>You can achieve the same inference results without using the pipeline, however, more steps will be required.</p> <p>Load the model from the 🤗 Hub:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>model = SpeechT5ForTextToSpeech.from_pretrained(<span class="hljs-string">"YOUR_ACCOUNT/speecht5_finetuned_voxpopuli_nl"</span>)<!----></pre></div><!----> <p>Pick an example from the test dataset and obtain a speaker embedding.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>example = dataset[<span class="hljs-string">"test"</span>][<span class="hljs-number">304</span>] | |
| <span class="hljs-meta">>>> </span>speaker_embeddings = torch.tensor(example[<span class="hljs-string">"speaker_embeddings"</span>]).unsqueeze(<span class="hljs-number">0</span>)<!----></pre></div><!----> <p>Define the input text and tokenize it.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>text = <span class="hljs-string">"hallo allemaal, ik praat nederlands. groetjes aan iedereen!"</span> | |
| <span class="hljs-meta">>>> </span>inputs = processor(text=text, return_tensors=<span class="hljs-string">"pt"</span>)<!----></pre></div><!----> <p>Create a spectrogram with your model:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>spectrogram = model.generate_speech(inputs[<span class="hljs-string">"input_ids"</span>], speaker_embeddings)<!----></pre></div><!----> <p>Visualize the spectrogram, if you’d like to:</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span>plt.figure() | |
| <span class="hljs-meta">>>> </span>plt.imshow(spectrogram.T) | |
| <span class="hljs-meta">>>> </span>plt.show()<!----></pre></div><!----> <div class="flex justify-center"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/transformers/tasks/tts_logmelspectrogram_2.png" alt="Generated log-mel spectrogram"/></div> <p>Finally, use the vocoder to turn the spectrogram into sound.</p> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg><!----> <div class=" absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0 "><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent;"></div> Copied</div><!----></button><!----></div> <pre class="language-py "><!----><span class="hljs-meta">>>> </span><span class="hljs-keyword">with</span> torch.no_grad(): | |
| <span class="hljs-meta">... </span> speech = vocoder(spectrogram) | |
| <span class="hljs-meta">>>> </span><span class="hljs-keyword">from</span> IPython.display <span class="hljs-keyword">import</span> Audio | |
| <span class="hljs-meta">>>> </span>Audio(speech.numpy(), rate=<span class="hljs-number">16000</span>)<!----></pre></div><!----> <p>In our experience, obtaining satisfactory results from this model can be challenging. The quality of the speaker | |
| embeddings appears to be a significant factor. Since SpeechT5 was pre-trained with English x-vectors, it performs best | |
| when using English speaker embeddings. If the synthesized speech sounds poor, try using a different speaker embedding.</p> <p>Increasing the training duration is also likely to enhance the quality of the results. Even so, the speech clearly is Dutch instead of English, and it does | |
| capture the voice characteristics of the speaker (compare to the original audio in the example). | |
| Another thing to experiment with is the model’s configuration. For example, try using <code>config.reduction_factor = 1</code> to | |
| see if this improves the results.</p> <p>Finally, it is essential to consider ethical considerations. Although TTS technology has numerous useful applications, it | |
| may also be used for malicious purposes, such as impersonating someone’s voice without their knowledge or consent. Please | |
| use TTS judiciously and responsibly.</p> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/transformers/blob/main/docs/source/en/tasks/text-to-speech.md" target="_blank"><svg class="mr-1" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M31,16l-7,7l-1.41-1.41L28.17,16l-5.58-5.59L24,9l7,7z"></path><path d="M1,16l7-7l1.41,1.41L3.83,16l5.58,5.59L8,23l-7-7z"></path><path d="M12.419,25.484L17.639,6.552l1.932,0.518L14.351,26.002z"></path></svg><!----> <span><span class="underline">Update</span> on GitHub</span></a><!----> <p></p><!--]--><!----><!--]--><!--]--><!--]--> <!--[-1--><!--]--><!--]--> | |
| <script> | |
| { | |
| __sveltekit_10jdx26 = { | |
| base: "/docs/transformers/main/en", | |
| assets: "/docs/transformers/main/en" | |
| }; | |
| const element = document.currentScript.parentElement; | |
| Promise.all([ | |
| import("/docs/transformers/main/en/_app/immutable/entry/start.QSJpda4J.js"), | |
| import("/docs/transformers/main/en/_app/immutable/entry/app.Ch9T9LuJ.js") | |
| ]).then(([kit, app]) => { | |
| kit.start(app, element, { | |
| node_ids: [0, 71], | |
| data: [null,null], | |
| form: null, | |
| error: null | |
| }); | |
| }); | |
| } | |
| </script> | |
Xet Storage Details
- Size:
- 127 kB
- Xet hash:
- caccb307ce23e50e7595816003d6da1c89db22dd147742bb0560805d6f30898f
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.