Update app.py

app.py

@@ -1,340 +1,204 @@
-from functools import lru_cache
+import logging
+import os
+import time
+import uuid
 
-import sherpa_onnx
-from huggingface_hub import hf_hub_download
+import gradio as gr
+import soundfile as sf
 
+from model import get_pretrained_model, language_to_models
 
-def get_file(
-    repo_id: str,
-    filename: str,
-    subfolder: str = ".",
-) -> str:
-    model_filename = hf_hub_download(
-        repo_id=repo_id,
-        filename=filename,
-        subfolder=subfolder,
-    )
-    return model_filename
+title = "Text-to-speech (TTS)-by-Haseeb Ahmed"
 
+description = """
+This space shows how to convert text to speech with Next-gen Kaldi.
 
-@lru_cache(maxsize=10)
-def _get_vits_vctk(repo_id: str, speed: float) -> sherpa_onnx.OfflineTts:
-    assert repo_id == "csukuangfj/vits-vctk"
+It is running on CPU within a docker container provided by Hugging Face.
 
-    model = get_file(
-        repo_id=repo_id,
-        filename="vits-vctk.onnx",
-        subfolder=".",
-    )
+"""
 
-    lexicon = get_file(
-        repo_id=repo_id,
-        filename="lexicon.txt",
-        subfolder=".",
-    )
+# css style is copied from
+# https://huggingface.co/spaces/alphacep/asr/blob/main/app.py#L113
+css = """
+.result {display:flex;flex-direction:column}
+.result_item {padding:15px;margin-bottom:8px;border-radius:15px;width:100%}
+.result_item_success {background-color:mediumaquamarine;color:white;align-self:start}
+.result_item_error {background-color:#ff7070;color:white;align-self:start}
+"""
+
+#examples = [
+#    ["Min-nan (闽南话)", "csukuangfj/vits-mms-nan", "ài piaǸ chiah ē iaN̂", 0, 1.0],
+#   ["Thai", "csukuangfj/vits-mms-tha", "ฉันรักคุณ", 0, 1.0],
+#]
 
-    tokens = get_file(
-        repo_id=repo_id,
-        filename="tokens.txt",
-        subfolder=".",
-    )
 
-    tts_config = sherpa_onnx.OfflineTtsConfig(
-        model=sherpa_onnx.OfflineTtsModelConfig(
-            vits=sherpa_onnx.OfflineTtsVitsModelConfig(
-                model=model,
-                lexicon=lexicon,
-                tokens=tokens,
-                length_scale=1.0 / speed,
-            ),
-            provider="cpu",
-            debug=True,
-            num_threads=2,
+def update_model_dropdown(language: str):
+    if language in language_to_models:
+        choices = language_to_models[language]
+        return gr.Dropdown(
+            choices=choices,
+            value=choices[0],
+            interactive=True,
         )
-    )
-    tts = sherpa_onnx.OfflineTts(tts_config)
 
-    return tts
+    raise ValueError(f"Unsupported language: {language}")
 
 
-@lru_cache(maxsize=10)
-def _get_vits_ljs(repo_id: str, speed: float) -> sherpa_onnx.OfflineTts:
-    assert repo_id == "csukuangfj/vits-ljs"
+def build_html_output(s: str, style: str = "result_item_success"):
+    return f"""
+    <div class='result'>
+        <div class='result_item {style}'>
+          {s}
+        </div>
+    </div>
+    """
 
-    model = get_file(
-        repo_id=repo_id,
-        filename="vits-ljs.onnx",
-        subfolder=".",
-    )
 
-    lexicon = get_file(
-        repo_id=repo_id,
-        filename="lexicon.txt",
-        subfolder=".",
-    )
+def process(language: str, repo_id: str, text: str, sid: str, speed: float):
+    logging.info(f"Input text: {text}. sid: {sid}, speed: {speed}")
+    sid = int(sid)
+    tts = get_pretrained_model(repo_id, speed)
 
-    tokens = get_file(
-        repo_id=repo_id,
-        filename="tokens.txt",
-        subfolder=".",
-    )
+    start = time.time()
+    audio = tts.generate(text, sid=sid)
+    end = time.time()
 
-    tts_config = sherpa_onnx.OfflineTtsConfig(
-        model=sherpa_onnx.OfflineTtsModelConfig(
-            vits=sherpa_onnx.OfflineTtsVitsModelConfig(
-                model=model,
-                lexicon=lexicon,
-                tokens=tokens,
-                length_scale=1.0 / speed,
-            ),
-            provider="cpu",
-            debug=True,
-            num_threads=2,
+    if len(audio.samples) == 0:
+        raise ValueError(
+            "Error in generating audios. Please read previous error messages."
         )
-    )
-    tts = sherpa_onnx.OfflineTts(tts_config)
-
-    return tts
-
-
-@lru_cache(maxsize=10)
-def _get_vits_piper(repo_id: str, speed: float) -> sherpa_onnx.OfflineTts:
-    data_dir = "/tmp/espeak-ng-data"
-    if "coqui" in repo_id or "vits-mms" in repo_id:
-        name = "model"
-    elif "piper" in repo_id:
-        n = len("vits-piper-")
-        name = repo_id.split("/")[1][n:]
-    elif "mimic3" in repo_id:
-        n = len("vits-mimic3-")
-        name = repo_id.split("/")[1][n:]
-    else:
-        raise ValueError(f"Unsupported {repo_id}")
-
-    if "vits-coqui-uk-mai" in repo_id or "vits-mms" in repo_id:
-        data_dir = ""
-
-    model = get_file(
-        repo_id=repo_id,
-        filename=f"{name}.onnx",
-        subfolder=".",
-    )
 
-    tokens = get_file(
-        repo_id=repo_id,
-        filename="tokens.txt",
-        subfolder=".",
-    )
+    duration = len(audio.samples) / audio.sample_rate
 
-    tts_config = sherpa_onnx.OfflineTtsConfig(
-        model=sherpa_onnx.OfflineTtsModelConfig(
-            vits=sherpa_onnx.OfflineTtsVitsModelConfig(
-                model=model,
-                lexicon="",
-                data_dir=data_dir,
-                tokens=tokens,
-                length_scale=1.0 / speed,
-            ),
-            provider="cpu",
-            debug=True,
-            num_threads=2,
+    elapsed_seconds = end - start
+    rtf = elapsed_seconds / duration
+
+    info = f"""
+    Wave duration  : {duration:.3f} s <br/>
+    Processing time: {elapsed_seconds:.3f} s <br/>
+    RTF: {elapsed_seconds:.3f}/{duration:.3f} = {rtf:.3f} <br/>
+    """
+
+    logging.info(info)
+    logging.info(f"\nrepo_id: {repo_id}\ntext: {text}\nsid: {sid}\nspeed: {speed}")
+
+    filename = str(uuid.uuid4())
+    filename = f"{filename}.wav"
+    sf.write(
+        filename,
+        audio.samples,
+        samplerate=audio.sample_rate,
+        subtype="PCM_16",
+    )
+
+    return filename, build_html_output(info)
+
+
+demo = gr.Blocks(css=css)
+
+
+with demo:
+    gr.Markdown(title)
+    language_choices = list(language_to_models.keys())
+
+    language_radio = gr.Radio(
+        label="Language",
+        choices=language_choices,
+        value=language_choices[0],
+    )
+
+    model_dropdown = gr.Dropdown(
+        choices=language_to_models[language_choices[0]],
+        label="Select a model",
+        value=language_to_models[language_choices[0]][0],
+    )
+
+    language_radio.change(
+        update_model_dropdown,
+        inputs=language_radio,
+        outputs=model_dropdown,
+    )
+
+    with gr.Tabs():
+        with gr.TabItem("Please input your text"):
+            input_text = gr.Textbox(
+                label="Input text",
+                info="Your text",
+                lines=3,
+                placeholder="Please input your text here",
+            )
+
+           # input_sid = gr.Textbox(
+            #    label="Speaker ID",
+             #   info="Speaker ID",
+              #  lines=1,
+               # max_lines=1,
+               # value="0",
+               # placeholder="Speaker ID. Valid only for mult-speaker model",
+            #)
+
+           # input_speed = gr.Slider(
+            #    minimum=0.1,
+             #   maximum=10,
+              #  value=1,
+               # step=0.1,
+                #label="Speed (larger->faster; smaller->slower)",
+            #)
+
+            input_button = gr.Button("Submit")
+
+            output_audio = gr.Audio(label="Output")
+
+            output_info = gr.HTML(label="Info")
+
+            #gr.Examples(
+               # examples=examples,
+               # fn=process,
+               # inputs=[
+                #    language_radio,
+                #    model_dropdown,
+                #    input_text,
+                #    input_sid,
+             #       input_speed,
+              #  ],
+              #  outputs=[
+               #     output_audio,
+               #     output_info,
+              #  ],
+          #  )
+
+        input_button.click(
+            process,
+            inputs=[
+                language_radio,
+                model_dropdown,
+                input_text,
+                #input_sid,
+                #input_speed,
+            ],
+            outputs=[
+                output_audio,
+                output_info,
+            ],
         )
-    )
-    tts = sherpa_onnx.OfflineTts(tts_config)
 
-    return tts
+    gr.Markdown(description)
 
 
-@lru_cache(maxsize=10)
-def _get_vits_mms(repo_id: str, speed: float) -> sherpa_onnx.OfflineTts:
-    return _get_vits_piper(repo_id, speed)
-
-
-@lru_cache(maxsize=10)
-def _get_vits_zh_aishell3(repo_id: str, speed: float) -> sherpa_onnx.OfflineTts:
-    assert repo_id == "csukuangfj/vits-zh-aishell3"
-
-    model = get_file(
-        repo_id=repo_id,
-        filename="vits-aishell3.onnx",
-        subfolder=".",
+def download_espeak_ng_data():
+    os.system(
+        """
+    cd /tmp
+    wget -qq https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/espeak-ng-data.tar.bz2
+    tar xf espeak-ng-data.tar.bz2
+    """
     )
 
-    lexicon = get_file(
-        repo_id=repo_id,
-        filename="lexicon.txt",
-        subfolder=".",
-    )
 
-    tokens = get_file(
-        repo_id=repo_id,
-        filename="tokens.txt",
-        subfolder=".",
-    )
-
-    rule_fst = get_file(
-        repo_id=repo_id,
-        filename="rule.fst",
-        subfolder=".",
-    )
-
-    tts_config = sherpa_onnx.OfflineTtsConfig(
-        model=sherpa_onnx.OfflineTtsModelConfig(
-            vits=sherpa_onnx.OfflineTtsVitsModelConfig(
-                model=model,
-                lexicon=lexicon,
-                tokens=tokens,
-                length_scale=1.0 / speed,
-            ),
-            provider="cpu",
-            debug=True,
-            num_threads=2,
-        ),
-        rule_fsts=rule_fst,
-    )
-    tts = sherpa_onnx.OfflineTts(tts_config)
-
-    return tts
-
-
-@lru_cache(maxsize=10)
-def _get_vits_hf(repo_id: str, speed: float) -> sherpa_onnx.OfflineTts:
-    if "fanchen" in repo_id or "vits-cantonese-hf-xiaomaiiwn" in repo_id:
-        model = repo_id.split("/")[-1]
-    else:
-        model = repo_id.split("-")[-1]
+if __name__ == "__main__":
+    download_espeak_ng_data()
+    formatter = "%(asctime)s %(levelname)s [%(filename)s:%(lineno)d] %(message)s"
 
-    model = get_file(
-        repo_id=repo_id,
-        filename=f"{model}.onnx",
-        subfolder=".",
-    )
-
-    lexicon = get_file(
-        repo_id=repo_id,
-        filename="lexicon.txt",
-        subfolder=".",
-    )
-
-    tokens = get_file(
-        repo_id=repo_id,
-        filename="tokens.txt",
-        subfolder=".",
-    )
-
-    rule_fst = get_file(
-        repo_id=repo_id,
-        filename="rule.fst",
-        subfolder=".",
-    )
+    logging.basicConfig(format=formatter, level=logging.INFO)
 
-    tts_config = sherpa_onnx.OfflineTtsConfig(
-        model=sherpa_onnx.OfflineTtsModelConfig(
-            vits=sherpa_onnx.OfflineTtsVitsModelConfig(
-                model=model,
-                lexicon=lexicon,
-                tokens=tokens,
-                length_scale=1.0 / speed,
-            ),
-            provider="cpu",
-            debug=True,
-            num_threads=2,
-        ),
-        rule_fsts=rule_fst,
-    )
-    tts = sherpa_onnx.OfflineTts(tts_config)
-
-    return tts
-
-
-@lru_cache(maxsize=10)
-def get_pretrained_model(repo_id: str, speed: float) -> sherpa_onnx.OfflineTts:
-    if repo_id in english_models:
-        return english_models[repo_id](repo_id, speed)
-    elif repo_id in arabic_models:
-        return arabic_models[repo_id](repo_id, speed)
-    elif repo_id in turkish_models:
-        return turkish_models[repo_id](repo_id, speed)
-    elif repo_id in persian_models:
-        return persian_models[repo_id](repo_id, speed)
-    elif repo_id in hindi_models:
-        return hindi_models[repo_id](repo_id, speed)
-    elif repo_id in gujarati_models:
-        return gujarati_models[repo_id](repo_id, speed)
-    else:
-        raise ValueError(f"Unsupported repo_id: {repo_id}")
-
-
-english_models = {
-    "csukuangfj/vits-piper-en_US-glados": _get_vits_piper,
-    # coqui-ai
-    "csukuangfj/vits-coqui-en-ljspeech": _get_vits_piper,
-    "csukuangfj/vits-coqui-en-ljspeech-neon": _get_vits_piper,
-    "csukuangfj/vits-coqui-en-vctk": _get_vits_piper,
-    # piper, US
-    "csukuangfj/vits-piper-en_GB-sweetbbak-amy": _get_vits_piper,
-    "csukuangfj/vits-piper-en_US-amy-low": _get_vits_piper,
-    "csukuangfj/vits-piper-en_US-amy-medium": _get_vits_piper,
-    "csukuangfj/vits-piper-en_US-arctic-medium": _get_vits_piper,  # 18 speakers
-    "csukuangfj/vits-piper-en_US-danny-low": _get_vits_piper,
-    "csukuangfj/vits-piper-en_US-hfc_male-medium": _get_vits_piper,
-    "csukuangfj/vits-piper-en_US-joe-medium": _get_vits_piper,
-    "csukuangfj/vits-piper-en_US-kathleen-low": _get_vits_piper,
-    "csukuangfj/vits-piper-en_US-kusal-medium": _get_vits_piper,
-    "csukuangfj/vits-piper-en_US-l2arctic-medium": _get_vits_piper,  # 24 speakers
-    "csukuangfj/vits-piper-en_US-lessac-low": _get_vits_piper,
-    "csukuangfj/vits-piper-en_US-lessac-medium": _get_vits_piper,
-    "csukuangfj/vits-piper-en_US-lessac-high": _get_vits_piper,
-    "csukuangfj/vits-piper-en_US-libritts-high": _get_vits_piper,  # 904 speakers
-    "csukuangfj/vits-piper-en_US-libritts_r-medium": _get_vits_piper,  # 904 speakers
-    "csukuangfj/vits-piper-en_US-ryan-low": _get_vits_piper,
-    "csukuangfj/vits-piper-en_US-ryan-medium": _get_vits_piper,
-    "csukuangfj/vits-piper-en_US-ryan-high": _get_vits_piper,
-    # piper, GB
-    "csukuangfj/vits-piper-en_GB-alan-low": _get_vits_piper,
-    "csukuangfj/vits-piper-en_GB-alan-medium": _get_vits_piper,
-    "csukuangfj/vits-piper-en_GB-alba-medium": _get_vits_piper,
-    "csukuangfj/vits-piper-en_GB-jenny_dioco-medium": _get_vits_piper,
-    "csukuangfj/vits-piper-en_GB-northern_english_male-medium": _get_vits_piper,
-    "csukuangfj/vits-piper-en_GB-semaine-medium": _get_vits_piper,
-    "csukuangfj/vits-piper-en_GB-southern_english_female-low": _get_vits_piper,
-    "csukuangfj/vits-piper-en_GB-vctk-medium": _get_vits_piper,
-    #
-    "csukuangfj/vits-vctk": _get_vits_vctk,  # 109 speakers
-    "csukuangfj/vits-ljs": _get_vits_ljs,
-}
-
-
-arabic_models = {
-    "csukuangfj/vits-piper-ar_JO-kareem-low": _get_vits_piper,
-    "csukuangfj/vits-piper-ar_JO-kareem-medium": _get_vits_piper,
-}
-
-turkish_models = {
-    "csukuangfj/vits-piper-tr_TR-dfki-medium": _get_vits_piper,
-    "csukuangfj/vits-piper-tr_TR-fahrettin-medium": _get_vits_piper,
-}
-
-
-persian_models = {
-    "csukuangfj/vits-piper-fa_IR-amir-medium": _get_vits_piper,
-    "csukuangfj/vits-piper-fa_IR-gyro-medium": _get_vits_piper,
-    "csukuangfj/vits-mimic3-fa-haaniye_low": _get_vits_piper,
-}
-
-gujarati_models = {
-    "csukuangfj/vits-mimic3-gu_IN-cmu-indic_low": _get_vits_piper,
-}
-hindi_models = {
-    "vosk-model-hi-0.22": _get_vits_piper,
-}
-
-
-language_to_models = {
-    "English": list(english_models.keys()),
-    "Arabic": list(arabic_models.keys()),
-    "Hindi": list(hindi_models.keys()),
-    "Gujarati": list(gujarati_models.keys()),
-    "Persian": list(persian_models.keys()),
-    "Turkish": list(turkish_models.keys()),
-}
+    demo.launch()