add multi model option

app.py

@@ -1 +1 @@
-run_demo.py
+run_demo_multi_models.py

run_demo_multi_models.py

@@ -0,0 +1,138 @@
+import logging
+
+import gradio as gr
+import pytube as pt
+import torch
+from huggingface_hub import model_info
+from transformers import pipeline
+
+DEFAULT_MODEL_NAME = "bhuang/whisper-medium-cv11-french-case-punctuation"
+MODEL_NAMES = [
+    "bhuang/whisper-small-cv11-french",
+    "bhuang/whisper-small-cv11-french-case-punctuation",
+    "bhuang/whisper-medium-cv11-french",
+    "bhuang/whisper-medium-cv11-french-case-punctuation",
+]
+CHUNK_LENGTH_S = 30
+
+logging.basicConfig(
+    format="%(asctime)s [%(levelname)s] [%(name)s] %(message)s",
+    datefmt="%Y-%m-%dT%H:%M:%SZ",
+)
+logger = logging.getLogger(__name__)
+logger.setLevel(logging.DEBUG)
+
+device = 0 if torch.cuda.is_available() else "cpu"
+
+cached_models = {}
+
+def maybe_load_cached_pipeline(model_name):
+    pipe = cached_models.get(model_name)
+    if pipe is None:
+        # load pipeline
+        pipe = pipeline(
+            task="automatic-speech-recognition",
+            model=model_name,
+            chunk_length_s=CHUNK_LENGTH_S,
+            device=device,
+        )
+        # set forced_decoder_ids
+        pipe.model.config.forced_decoder_ids = pipe.tokenizer.get_decoder_prompt_ids(language="fr", task="transcribe")
+
+        logger.info(f"`{model_name}` pipeline has been initialized")
+
+        cached_models[model_name] = pipe
+    return pipe
+
+
+def transcribe(microphone, file_upload, model_name):
+    warn_output = ""
+    if (microphone is not None) and (file_upload is not None):
+        warn_output = (
+            "WARNING: You've uploaded an audio file and used the microphone. "
+            "The recorded file from the microphone will be used and the uploaded audio will be discarded.\n"
+        )
+
+    elif (microphone is None) and (file_upload is None):
+        return "ERROR: You have to either use the microphone or upload an audio file"
+
+    file = microphone if microphone is not None else file_upload
+
+    pipe = maybe_load_cached_pipeline(model_name)
+    text = pipe(file)["text"]
+
+    logger.info(f"Transcription: {text}")
+
+    return warn_output + text
+
+
+def _return_yt_html_embed(yt_url):
+    video_id = yt_url.split("?v=")[-1]
+    HTML_str = (
+        f'<center> <iframe width="500" height="320" src="https://www.youtube.com/embed/{video_id}"> </iframe>'
+        " </center>"
+    )
+    return HTML_str
+
+
+def yt_transcribe(yt_url, model_name):
+    yt = pt.YouTube(yt_url)
+    html_embed_str = _return_yt_html_embed(yt_url)
+    stream = yt.streams.filter(only_audio=True)[0]
+    stream.download(filename="audio.mp3")
+
+    pipe = maybe_load_cached_pipeline(model_name)
+    text = pipe("audio.mp3")["text"]
+
+    logger.info(f"Transcription: {text}")
+
+    return html_embed_str, text
+
+
+# load default model
+maybe_load_cached_pipeline(DEFAULT_MODEL_NAME)
+
+demo = gr.Blocks()
+
+mf_transcribe = gr.Interface(
+    fn=transcribe,
+    inputs=[
+        gr.inputs.Audio(source="microphone", type="filepath", optional=True),
+        gr.inputs.Audio(source="upload", type="filepath", optional=True),
+        gr.inputs.Dropdown(choices=MODEL_NAMES, default=DEFAULT_MODEL_NAME, label="Whisper Model"),
+    ],
+    outputs="text",
+    layout="horizontal",
+    theme="huggingface",
+    title="Whisper Demo: Transcribe Audio",
+    description=(
+        "Transcribe long-form microphone or audio inputs with the click of a button! Demo uses the the fine-tuned"
+        f" checkpoint [{DEFAULT_MODEL_NAME}](https://huggingface.co/{DEFAULT_MODEL_NAME}) and 🤗 Transformers to transcribe audio files"
+        " of arbitrary length."
+    ),
+    allow_flagging="never",
+)
+
+yt_transcribe = gr.Interface(
+    fn=yt_transcribe,
+    inputs=[
+        gr.inputs.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL"),
+        gr.inputs.Dropdown(choices=MODEL_NAMES, default=DEFAULT_MODEL_NAME, label="Whisper Model"),
+    ],
+    outputs=["html", "text"],
+    layout="horizontal",
+    theme="huggingface",
+    title="Whisper Demo: Transcribe YouTube",
+    description=(
+        "Transcribe long-form YouTube videos with the click of a button! Demo uses the the fine-tuned checkpoint:"
+        f" [{DEFAULT_MODEL_NAME}](https://huggingface.co/{DEFAULT_MODEL_NAME}) and 🤗 Transformers to transcribe audio files of"
+        " arbitrary length."
+    ),
+    allow_flagging="never",
+)
+
+with demo:
+    gr.TabbedInterface([mf_transcribe, yt_transcribe], ["Transcribe Audio", "Transcribe YouTube"])
+
+# demo.launch(server_name="0.0.0.0", debug=True, share=True)
+demo.launch(enable_queue=True)