Create app.py

app.py

@@ -0,0 +1,218 @@
+import torch
+
+import gradio as gr
+import yt_dlp as youtube_dl
+import numpy as np
+from datasets import Dataset, Audio
+from scipy.io import wavfile
+
+from transformers import pipeline
+from transformers.pipelines.audio_utils import ffmpeg_read
+
+import tempfile
+import os
+import time
+os.environ["GRADIO_TEMP_DIR"] = "/home/yoach/spaces/tmp"
+
+
+MODEL_NAME = "openai/whisper-large-v3"
+BATCH_SIZE = 8
+FILE_LIMIT_MB = 1000
+YT_LENGTH_LIMIT_S = 3600  # limit to 1 hour YouTube files
+
+device = 0 if torch.cuda.is_available() else "cpu"
+
+pipe = pipeline(
+    task="automatic-speech-recognition",
+    model=MODEL_NAME,
+    chunk_length_s=30,
+    device=device,
+)
+
+
+def transcribe(inputs_path, task, dataset_name, oauth_token: gr.OAuthToken):
+    if inputs_path is None:
+        raise gr.Error("No audio file submitted! Please upload or record an audio file before submitting your request.")
+    
+    sampling_rate, inputs = wavfile.read(inputs_path) 
+
+    out = pipe(inputs_path, batch_size=BATCH_SIZE, generate_kwargs={"task": task}, return_timestamps=True)
+    
+    text = out["text"]
+    
+    chunks = naive_postprocess_whisper_chunks(out["chunks"])
+        
+    transcripts = []
+    audios = []
+    with tempfile.TemporaryDirectory() as tmpdirname:
+        for i,chunk in enumerate(chunks):
+            begin, end = chunk["timestamp"]
+            begin, end = int(begin*sampling_rate), int(end*sampling_rate)
+            # TODO: make sure 1D or 2D?
+            arr = inputs[begin:end]
+            path = os.path.join(tmpdirname, f"{i}.wav")
+            wavfile.write(path, sampling_rate,  arr)
+            audios.append(path)
+            transcripts.append(chunk["text"])
+            
+        dataset = Dataset.from_dict({"audio": audios, "transcript": transcripts}).cast_column("audio", Audio())
+        
+    
+        dataset.push_to_hub(dataset_name, token=oauth_token)
+        
+    return  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 download_yt_audio(yt_url, filename):
+    info_loader = youtube_dl.YoutubeDL()
+    
+    try:
+        info = info_loader.extract_info(yt_url, download=False)
+    except youtube_dl.utils.DownloadError as err:
+        raise gr.Error(str(err))
+    
+    file_length = info["duration_string"]
+    file_h_m_s = file_length.split(":")
+    file_h_m_s = [int(sub_length) for sub_length in file_h_m_s]
+    
+    if len(file_h_m_s) == 1:
+        file_h_m_s.insert(0, 0)
+    if len(file_h_m_s) == 2:
+        file_h_m_s.insert(0, 0)
+    file_length_s = file_h_m_s[0] * 3600 + file_h_m_s[1] * 60 + file_h_m_s[2]
+    
+    if file_length_s > YT_LENGTH_LIMIT_S:
+        yt_length_limit_hms = time.strftime("%HH:%MM:%SS", time.gmtime(YT_LENGTH_LIMIT_S))
+        file_length_hms = time.strftime("%HH:%MM:%SS", time.gmtime(file_length_s))
+        raise gr.Error(f"Maximum YouTube length is {yt_length_limit_hms}, got {file_length_hms} YouTube video.")
+    
+    ydl_opts = {"outtmpl": filename, "format": "worstvideo[ext=mp4]+bestaudio[ext=m4a]/best[ext=mp4]/best"}
+    
+    with youtube_dl.YoutubeDL(ydl_opts) as ydl:
+        try:
+            ydl.download([yt_url])
+        except youtube_dl.utils.ExtractorError as err:
+            raise gr.Error(str(err))
+
+
+def yt_transcribe(yt_url, task, dataset_name, oauth_token: gr.OAuthToken, max_filesize=75.0, dataset_sampling_rate = 24000):
+    html_embed_str = _return_yt_html_embed(yt_url)
+
+    with tempfile.TemporaryDirectory() as tmpdirname:
+        filepath = os.path.join(tmpdirname, "video.mp4")
+        download_yt_audio(yt_url, filepath)
+        with open(filepath, "rb") as f:
+            inputs_path = f.read()
+
+    inputs = ffmpeg_read(inputs_path, pipe.feature_extractor.sampling_rate)
+    inputs = {"array": inputs, "sampling_rate": pipe.feature_extractor.sampling_rate}
+
+    out = pipe(inputs, batch_size=BATCH_SIZE, generate_kwargs={"task": task}, return_timestamps=True)
+    
+    text = out["text"]
+    
+    chunks = naive_postprocess_whisper_chunks(out["chunks"])
+    
+    inputs = ffmpeg_read(inputs_path, dataset_sampling_rate)
+    
+    transcripts = []
+    audios = []
+    with tempfile.TemporaryDirectory() as tmpdirname:
+        for i,chunk in enumerate(chunks):
+            begin, end = chunk["timestamp"]
+            begin, end = int(begin*dataset_sampling_rate), int(end*dataset_sampling_rate)
+            # TODO: make sure 1D or 2D?
+            arr = inputs[begin:end]
+            path = os.path.join(tmpdirname, f"{i}.wav")
+            wavfile.write(path, dataset_sampling_rate,  arr)
+            audios.append(path)
+            transcripts.append(chunk["text"])
+            
+        dataset = Dataset.from_dict({"audio": audios, "transcript": transcripts}).cast_column("audio", Audio())
+        
+    
+        dataset.push_to_hub(dataset_name, token=oauth_token)
+        
+
+    return html_embed_str, text
+
+
+def naive_postprocess_whisper_chunks(chunks, stop_chars = ".!:;?", min_duration = 5):
+    new_chunks = []
+    
+    while chunks:
+        current_chunk = chunks.pop(0)
+        begin, end = current_chunk["timestamp"]
+        text = current_chunk["text"]
+        
+        while chunks and (text[-1] not in stop_chars or (end-begin<min_duration)):
+            ch = chunks.pop(0)
+            end = ch["timestamp"][1]
+            text = "".join([text, ch["text"]])
+            
+        new_chunks.append({
+            "text": text.strip(),
+            "timestamp": (begin, end),
+        })
+        print(f"LENGTH CHUNK #{len(new_chunks)}: {end-begin}s")
+            
+    return new_chunks
+    
+    
+    
+    
+    
+
+demo = gr.Blocks()
+
+mf_transcribe = gr.Interface(
+    fn=transcribe,
+    inputs=[
+        gr.Audio(type="filepath"),
+        gr.Radio(["transcribe", "translate"], label="Task", value="transcribe"),
+        gr.Textbox(lines=1, placeholder="Place your new dataset name here", label="Dataset name"),
+    ],
+    outputs="text",
+    theme="huggingface",
+    title="Whisper Large V3: Transcribe Audio",
+    description=(
+        "Transcribe long-form microphone or audio inputs with the click of a button! Demo uses the"
+        f" checkpoint [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and 🤗 Transformers to transcribe audio files"
+        " of arbitrary length."
+    ),
+    allow_flagging="never",
+)
+
+yt_transcribe = gr.Interface(
+    fn=yt_transcribe,
+    inputs=[
+        gr.Textbox(lines=1, placeholder="Paste the URL to a YouTube video here", label="YouTube URL"),
+        gr.Radio(["transcribe", "translate"], label="Task", value="transcribe"),
+        gr.Textbox(lines=1, placeholder="Place your new dataset name here", label="Dataset name"),
+    ],
+    outputs=["html", "text"],
+    theme="huggingface",
+    title="Whisper Large V3: Transcribe YouTube",
+    description=(
+        "Transcribe long-form YouTube videos with the click of a button! Demo uses the checkpoint"
+        f" [{MODEL_NAME}](https://huggingface.co/{MODEL_NAME}) and 🤗 Transformers to transcribe video files of"
+        " arbitrary length."
+    ),
+    allow_flagging="never",
+)
+
+with demo:
+    with gr.Row():
+        gr.LoginButton()
+        gr.LogoutButton()
+    gr.TabbedInterface([mf_transcribe, yt_transcribe], ["Microphone or Audio file", "YouTube"])
+
+demo.launch(debug=True)