Buckets:
| <meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Preprocessing an audio dataset","local":"preprocessing-an-audio-dataset","sections":[{"title":"Resampling the audio data","local":"resampling-the-audio-data","sections":[],"depth":2},{"title":"Filtering the dataset","local":"filtering-the-dataset","sections":[],"depth":2},{"title":"Pre-processing audio data","local":"pre-processing-audio-data","sections":[],"depth":2}],"depth":1}"> | |
| <link href="/docs/audio-course/pr_201/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload"> | |
| <link rel="modulepreload" href="/docs/audio-course/pr_201/en/_app/immutable/entry/start.367c4d78.js"> | |
| <link rel="modulepreload" href="/docs/audio-course/pr_201/en/_app/immutable/chunks/scheduler.f7e1785c.js"> | |
| <link rel="modulepreload" href="/docs/audio-course/pr_201/en/_app/immutable/chunks/singletons.0d70d4cc.js"> | |
| <link rel="modulepreload" href="/docs/audio-course/pr_201/en/_app/immutable/chunks/index.279db187.js"> | |
| <link rel="modulepreload" href="/docs/audio-course/pr_201/en/_app/immutable/chunks/paths.274f629d.js"> | |
| <link rel="modulepreload" href="/docs/audio-course/pr_201/en/_app/immutable/entry/app.4c54ebf9.js"> | |
| <link rel="modulepreload" href="/docs/audio-course/pr_201/en/_app/immutable/chunks/index.9f8f0838.js"> | |
| <link rel="modulepreload" href="/docs/audio-course/pr_201/en/_app/immutable/nodes/0.e329f606.js"> | |
| <link rel="modulepreload" href="/docs/audio-course/pr_201/en/_app/immutable/chunks/each.e59479a4.js"> | |
| <link rel="modulepreload" href="/docs/audio-course/pr_201/en/_app/immutable/nodes/8.be3a7ddb.js"> | |
| <link rel="modulepreload" href="/docs/audio-course/pr_201/en/_app/immutable/chunks/Tip.4575d9cf.js"> | |
| <link rel="modulepreload" href="/docs/audio-course/pr_201/en/_app/immutable/chunks/CodeBlock.b3510e34.js"> | |
| <link rel="modulepreload" href="/docs/audio-course/pr_201/en/_app/immutable/chunks/EditOnGithub.5a9bb8c5.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"Preprocessing an audio dataset","local":"preprocessing-an-audio-dataset","sections":[{"title":"Resampling the audio data","local":"resampling-the-audio-data","sections":[],"depth":2},{"title":"Filtering the dataset","local":"filtering-the-dataset","sections":[],"depth":2},{"title":"Pre-processing audio data","local":"pre-processing-audio-data","sections":[],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <h1 class="relative group"><a id="preprocessing-an-audio-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="#preprocessing-an-audio-dataset"><span><svg class="" 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>Preprocessing an audio dataset</span></h1> <p data-svelte-h="svelte-1b3xnqq">Loading a dataset with 🤗 Datasets is just half of the fun. If you plan to use it either for training a model, or for running | |
| inference, you will need to pre-process the data first. In general, this will involve the following steps:</p> <ul data-svelte-h="svelte-1anvx2"><li>Resampling the audio data</li> <li>Filtering the dataset</li> <li>Converting audio data to model’s expected input</li></ul> <h2 class="relative group"><a id="resampling-the-audio-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="#resampling-the-audio-data"><span><svg class="" 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>Resampling the audio data</span></h2> <p data-svelte-h="svelte-47dy14">The <code>load_dataset</code> function downloads audio examples with the sampling rate that they were published with. This is not | |
| always the sampling rate expected by a model you plan to train, or use for inference. If there’s a discrepancy between | |
| the sampling rates, you can resample the audio to the model’s expected sampling rate.</p> <p data-svelte-h="svelte-1c2c4da">Most of the available pretrained models have been pretrained on audio datasets at a sampling rate of 16 kHz. | |
| When we explored MINDS-14 dataset, you may have noticed that it is sampled at 8 kHz, which means we will likely need | |
| to upsample it.</p> <p data-svelte-h="svelte-x7frc7">To do so, use 🤗 Datasets’ <code>cast_column</code> method. This operation does not change the audio in-place, but rather signals | |
| to datasets to resample the audio examples on the fly when they are loaded. The following code will set the sampling | |
| rate to 16kHz:</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 class="" 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=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> datasets <span class="hljs-keyword">import</span> Audio | |
| minds = minds.cast_column(<span class="hljs-string">"audio"</span>, Audio(sampling_rate=<span class="hljs-number">16_000</span>))<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-96ugis">Re-load the first audio example in the MINDS-14 dataset, and check that it has been resampled to the desired <code>sampling rate</code>:</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 class="" 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=""><!-- HTML_TAG_START -->minds[<span class="hljs-number">0</span>]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1mvdyro"><strong>Output:</strong></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 class="" 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=""><!-- HTML_TAG_START -->{ | |
| <span class="hljs-comment">"path"</span>: <span class="hljs-comment">"/root/.cache/huggingface/datasets/downloads/extracted/f14948e0e84be638dd7943ac36518a4cf3324e8b7aa331c5ab11541518e9368c/en-AU~PAY_BILL/response_4.wav"</span>, | |
| <span class="hljs-comment">"audio"</span>: { | |
| <span class="hljs-comment">"path"</span>: <span class="hljs-comment">"/root/.cache/huggingface/datasets/downloads/extracted/f14948e0e84be638dd7943ac36518a4cf3324e8b7aa331c5ab11541518e9368c/en-AU~PAY_BILL/response_4.wav"</span>, | |
| <span class="hljs-comment">"array"</span>: array( | |
| [ | |
| <span class="hljs-number">2.0634243e-05</span>, | |
| <span class="hljs-number">1.9437837e-04</span>, | |
| <span class="hljs-number">2.2419340e-04</span>, | |
| ..., | |
| <span class="hljs-number">9.3852862e-04</span>, | |
| <span class="hljs-number">1.1302452e-03</span>, | |
| <span class="hljs-number">7.1531429e-04</span>, | |
| ], | |
| dtype=float32, | |
| ), | |
| <span class="hljs-comment">"sampling_rate"</span>: <span class="hljs-number">16000</span>, | |
| }, | |
| <span class="hljs-comment">"transcription"</span>: <span class="hljs-comment">"I would like to pay my electricity bill using my card can you please assist"</span>, | |
| <span class="hljs-comment">"intent_class"</span>: <span class="hljs-number">13</span>, | |
| }<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-hnb1k">You may notice that the array values are now also different. This is because we’ve now got twice the number of amplitude values for | |
| every one that we had before.</p> <div class="course-tip bg-gradient-to-br dark:bg-gradient-to-r before:border-green-500 dark:before:border-green-800 from-green-50 dark:from-gray-900 to-white dark:to-gray-950 border border-green-50 text-green-700 dark:text-gray-400">💡 Some background on resampling: If an audio signal has been sampled at 8 kHz, so that it has 8000 sample readings per | |
| second, we know that the audio does not contain any frequencies over 4 kHz. This is guaranteed by the Nyquist sampling | |
| theorem. Because of this, we can be certain that in between the sampling points the original continuous signal always | |
| makes a smooth curve. Upsampling to a higher sampling rate is then a matter of calculating additional sample values that go in between | |
| the existing ones, by approximating this curve. Downsampling, however, requires that we first filter out any frequencies | |
| that would be higher than the new Nyquist limit, before estimating the new sample points. In other words, you can't | |
| downsample by a factor 2x by simply throwing away every other sample — this will create distortions in the signal called | |
| aliases. Doing resampling correctly is tricky and best left to well-tested libraries such as librosa or 🤗 Datasets.</div> <h2 class="relative group"><a id="filtering-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="#filtering-the-dataset"><span><svg class="" 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>Filtering the dataset</span></h2> <p data-svelte-h="svelte-s6y06a">You may need to filter the data based on some criteria. One of the common cases involves limiting the audio examples to a | |
| certain duration. For instance, we might want to filter out any examples longer than 20s to prevent out-of-memory errors | |
| when training a model.</p> <p data-svelte-h="svelte-1szy3nk">We can do this by using the 🤗 Datasets’ <code>filter</code> method and passing a function with filtering logic to it. Let’s start by writing a | |
| function that indicates which examples to keep and which to discard. This function, <code>is_audio_length_in_range</code>, | |
| returns <code>True</code> if a sample is shorter than 20s, and <code>False</code> if it is longer than 20s.</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 class="" 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=""><!-- HTML_TAG_START -->MAX_DURATION_IN_SECONDS = <span class="hljs-number">20.0</span> | |
| <span class="hljs-keyword">def</span> <span class="hljs-title function_">is_audio_length_in_range</span>(<span class="hljs-params">input_length</span>): | |
| <span class="hljs-keyword">return</span> input_length < MAX_DURATION_IN_SECONDS<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1btp9dp">The filtering function can be applied to a dataset’s column but we do not have a column with audio track duration in this | |
| dataset. However, we can create one, filter based on the values in that column, and then remove 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 class="" 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=""><!-- HTML_TAG_START --><span class="hljs-comment"># use librosa to get example's duration from the audio file</span> | |
| new_column = [librosa.get_duration(path=x) <span class="hljs-keyword">for</span> x <span class="hljs-keyword">in</span> minds[<span class="hljs-string">"path"</span>]] | |
| minds = minds.add_column(<span class="hljs-string">"duration"</span>, new_column) | |
| <span class="hljs-comment"># use 🤗 Datasets' `filter` method to apply the filtering function</span> | |
| minds = minds.<span class="hljs-built_in">filter</span>(is_audio_length_in_range, input_columns=[<span class="hljs-string">"duration"</span>]) | |
| <span class="hljs-comment"># remove the temporary helper column</span> | |
| minds = minds.remove_columns([<span class="hljs-string">"duration"</span>]) | |
| minds<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1mvdyro"><strong>Output:</strong></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 class="" 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=""><!-- HTML_TAG_START --><span class="hljs-constructor">Dataset({<span class="hljs-params">features</span>: [<span class="hljs-string">"path"</span>, <span class="hljs-string">"audio"</span>, <span class="hljs-string">"transcription"</span>, <span class="hljs-string">"intent_class"</span>], <span class="hljs-params">num_rows</span>: 624})</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-ipsrxo">We can verify that dataset has been filtered down from 654 examples to 624.</p> <h2 class="relative group"><a id="pre-processing-audio-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="#pre-processing-audio-data"><span><svg class="" 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>Pre-processing audio data</span></h2> <p data-svelte-h="svelte-jbw2e0">One of the most challenging aspects of working with audio datasets is preparing the data in the right format for model | |
| training. As you saw, the raw audio data comes as an array of sample values. However, pre-trained models, whether you use them | |
| for inference, or want to fine-tune them for your task, expect the raw data to be converted into input features. The | |
| requirements for the input features may vary from one model to another — they depend on the model’s architecture, and the data it was | |
| pre-trained with. The good news is, for every supported audio model, 🤗 Transformers offer a feature extractor class | |
| that can convert raw audio data into the input features the model expects.</p> <p data-svelte-h="svelte-1jndbwe">So what does a feature extractor do with the raw audio data? Let’s take a look at <a href="https://huggingface.co/papers/2212.04356" rel="nofollow">Whisper</a>’s | |
| feature extractor to understand some common feature extraction transformations. Whisper is a pre-trained model for | |
| automatic speech recognition (ASR) published in September 2022 by Alec Radford et al. from OpenAI.</p> <p data-svelte-h="svelte-1y8nv8f">First, the Whisper feature extractor pads/truncates a batch of audio examples such that all | |
| examples have an input length of 30s. Examples shorter than this are padded to 30s by appending zeros to the end of the | |
| sequence (zeros in an audio signal correspond to no signal or silence). Examples longer than 30s are truncated to 30s. | |
| Since all elements in the batch are padded/truncated to a maximum length in the input space, there is no need for an attention | |
| mask. Whisper is unique in this regard, most other audio models require an attention mask that details | |
| where sequences have been padded, and thus where they should be ignored in the self-attention mechanism. Whisper is | |
| trained to operate without an attention mask and infer directly from the speech signals where to ignore the inputs.</p> <p data-svelte-h="svelte-1ys9rww">The second operation that the Whisper feature extractor performs is converting the padded audio arrays to log-mel spectrograms. | |
| As you recall, these spectrograms describe how the frequencies of a signal change over time, expressed on the mel scale | |
| and measured in decibels (the log part) to make the frequencies and amplitudes more representative of human hearing.</p> <p data-svelte-h="svelte-n9ngno">All these transformations can be applied to your raw audio data with a couple of lines of code. Let’s go ahead and load | |
| the feature extractor from the pre-trained Whisper checkpoint to have ready for our audio 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 class="" 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=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> WhisperFeatureExtractor | |
| feature_extractor = WhisperFeatureExtractor.from_pretrained(<span class="hljs-string">"openai/whisper-small"</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-kd1gw8">Next, you can write a function to pre-process a single audio example by passing it through the <code>feature_extractor</code>.</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 class="" 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=""><!-- HTML_TAG_START --><span class="hljs-keyword">def</span> <span class="hljs-title function_">prepare_dataset</span>(<span class="hljs-params">example</span>): | |
| audio = example[<span class="hljs-string">"audio"</span>] | |
| features = feature_extractor( | |
| audio[<span class="hljs-string">"array"</span>], sampling_rate=audio[<span class="hljs-string">"sampling_rate"</span>], padding=<span class="hljs-literal">True</span> | |
| ) | |
| <span class="hljs-keyword">return</span> features<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-6ixft2">We can apply the data preparation function to all of our training examples using 🤗 Datasets’ map method:</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 class="" 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=""><!-- HTML_TAG_START -->minds = minds.<span class="hljs-built_in">map</span>(prepare_dataset) | |
| minds<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1mvdyro"><strong>Output:</strong></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 class="" 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=""><!-- HTML_TAG_START -->Dataset( | |
| <span class="hljs-punctuation">{</span> | |
| <span class="hljs-symbol"> features:</span> [<span class="hljs-string">"path"</span>, <span class="hljs-string">"audio"</span>, <span class="hljs-string">"transcription"</span>, <span class="hljs-string">"intent_class"</span>, <span class="hljs-string">"input_features"</span>], | |
| <span class="hljs-symbol"> num_rows:</span> <span class="hljs-number">624</span>, | |
| <span class="hljs-punctuation">}</span> | |
| )<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-14djzfz">As easy as that, we now have log-mel spectrograms as <code>input_features</code> in the dataset.</p> <p data-svelte-h="svelte-1478ups">Let’s visualize it for one of the examples in the <code>minds</code> 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 class="" 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=""><!-- HTML_TAG_START --><span class="hljs-keyword">import</span> numpy <span class="hljs-keyword">as</span> np | |
| example = minds[<span class="hljs-number">0</span>] | |
| input_features = example[<span class="hljs-string">"input_features"</span>] | |
| plt.figure().set_figwidth(<span class="hljs-number">12</span>) | |
| librosa.display.specshow( | |
| np.asarray(input_features[<span class="hljs-number">0</span>]), | |
| x_axis=<span class="hljs-string">"time"</span>, | |
| y_axis=<span class="hljs-string">"mel"</span>, | |
| sr=feature_extractor.sampling_rate, | |
| hop_length=feature_extractor.hop_length, | |
| ) | |
| plt.colorbar()<!-- HTML_TAG_END --></pre></div> <div class="flex justify-center" data-svelte-h="svelte-csckl"><img src="https://huggingface.co/datasets/huggingface-course/audio-course-images/resolve/main/log_mel_whisper.png" alt="Log mel spectrogram plot"></div> <p data-svelte-h="svelte-pj9vlt">Now you can see what the audio input to the Whisper model looks like after preprocessing.</p> <p data-svelte-h="svelte-1x1eerf">The model’s feature extractor class takes care of transforming raw audio data to the format that the model expects. However, | |
| many tasks involving audio are multimodal, e.g. speech recognition. In such cases 🤗 Transformers also offer model-specific | |
| tokenizers to process the text inputs. For a deep dive into tokenizers, please refer to our <a href="https://huggingface.co/course/chapter2/4" rel="nofollow">NLP course</a>.</p> <p data-svelte-h="svelte-ngi3x1">You can load the feature extractor and tokenizer for Whisper and other multimodal models separately, or you can load both via | |
| a so-called processor. To make things even simpler, use <code>AutoProcessor</code> to load a model’s feature extractor and processor from a | |
| checkpoint, like this:</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 class="" 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=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoProcessor | |
| processor = AutoProcessor.from_pretrained(<span class="hljs-string">"openai/whisper-small"</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-gvmmsi">Here we have illustrated the fundamental data preparation steps. Of course, custom data may require more complex preprocessing. | |
| In this case, you can extend the function <code>prepare_dataset</code> to perform any sort of custom data transformations. With 🤗 Datasets, | |
| if you can write it as a Python function, you can <a href="https://huggingface.co/docs/datasets/audio_process" rel="nofollow">apply it</a> to your dataset!</p> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/audio-transformers-course/blob/main/chapters/en/chapter1/preprocessing.mdx" target="_blank"><span data-svelte-h="svelte-1kd6by1"><</span> <span data-svelte-h="svelte-x0xyl0">></span> <span data-svelte-h="svelte-1dajgef"><span class="underline ml-1.5">Update</span> on GitHub</span></a> <p></p> | |
| <script> | |
| { | |
| __sveltekit_yq3w38 = { | |
| assets: "/docs/audio-course/pr_201/en", | |
| base: "/docs/audio-course/pr_201/en", | |
| env: {} | |
| }; | |
| const element = document.currentScript.parentElement; | |
| const data = [null,null]; | |
| Promise.all([ | |
| import("/docs/audio-course/pr_201/en/_app/immutable/entry/start.367c4d78.js"), | |
| import("/docs/audio-course/pr_201/en/_app/immutable/entry/app.4c54ebf9.js") | |
| ]).then(([kit, app]) => { | |
| kit.start(app, element, { | |
| node_ids: [0, 8], | |
| data, | |
| form: null, | |
| error: null | |
| }); | |
| }); | |
| } | |
| </script> | |
Xet Storage Details
- Size:
- 37.7 kB
- Xet hash:
- fbca38f5de3b8319008dea1e60b1de607f8ee1c3a71fb06b26c44dd58e50d34a
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.