initial upload

.gitignore

@@ -0,0 +1,2 @@
+myenv
+*.wav

app.py

@@ -0,0 +1,123 @@
+import torch
+from transformers import pipeline
+import os
+import gradio as gr
+from pydub import AudioSegment
+from pytube import YouTube
+import timeit
+import math
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+pipe = pipeline("automatic-speech-recognition", model="distil-whisper/distil-medium.en", device=device)
+
+def transcribe_speech_local(filepath):
+    if filepath is None:
+        return [{"error": "No audio found, please retry."}]
+
+    # Split audio into 15-second chunks
+    audio = AudioSegment.from_file(filepath)
+    chunk_length_ms = 15000  # 15 seconds in milliseconds
+    chunks = [audio[i:i + chunk_length_ms] for i in range(0, len(audio), chunk_length_ms)]
+    print(chunks)
+    aligned_chunks = []
+    transcription_time_total = 0
+
+    # Transcribe each chunk and measure time
+    for chunk_id, chunk in enumerate(chunks):
+        start_time = timeit.default_timer()
+        chunk.export("temp_chunk.wav", format="wav")
+        output = pipe("temp_chunk.wav")
+        transcription_time = timeit.default_timer() - start_time
+        transcription_time_total += transcription_time
+
+        # Calculate start and end times in seconds
+        start_time_sec = chunk_id * 15
+        end_time_sec = start_time_sec + len(chunk) / 1000.0
+
+        aligned_chunks.append({
+            "chunk_id": chunk_id,
+            "chunk_length": len(chunk) / 1000.0,
+            "text": output["text"],
+            "start_time": start_time_sec,
+            "end_time": end_time_sec,
+            "transcription_time": transcription_time
+        })
+
+    return aligned_chunks
+
+def download_audio_from_youtube(youtube_url):
+    yt = YouTube(youtube_url)
+    stream = yt.streams.filter(only_audio=True).first()
+    output_path = stream.download()
+    base, ext = os.path.splitext(output_path)
+    audio_file = base + '.mp3'
+    os.rename(output_path, audio_file)
+    return audio_file
+
+def transcribe_speech_from_youtube(youtube_url):
+    audio_filepath = download_audio_from_youtube(youtube_url)
+
+    # Convert to WAV format with 16kHz sample rate if necessary
+    audio = AudioSegment.from_file(audio_filepath)
+    audio = audio.set_frame_rate(16000).set_channels(1)
+    audio.export("converted_audio.wav", format="wav")
+    audio = AudioSegment.from_file("converted_audio.wav")
+
+    # Split audio into 15-second chunks
+    chunk_length_ms = 15000  # 15 seconds in milliseconds
+    chunks = [audio[i:i + chunk_length_ms] for i in range(0, len(audio), chunk_length_ms)]
+
+    aligned_chunks = []
+    transcription_time_total = 0
+
+    # Transcribe each chunk and measure time
+    for chunk_id, chunk in enumerate(chunks):
+        start_time = timeit.default_timer()
+        chunk.export("temp_chunk.wav", format="wav")
+        output = pipe("temp_chunk.wav")
+        transcription_time = timeit.default_timer() - start_time
+        transcription_time_total += transcription_time
+
+        # Calculate start and end times in seconds
+        start_time_sec = chunk_id * 15
+        end_time_sec = start_time_sec + len(chunk) / 1000.0
+
+        aligned_chunks.append({
+            "chunk_id": chunk_id,
+            "chunk_length": len(chunk) / 1000.0,
+            "text": output["text"],
+            "start_time": start_time_sec,
+            "end_time": end_time_sec,
+            "transcription_time": transcription_time
+        })
+
+    # Clean up temporary files
+    if os.path.exists("temp_chunk.wav"):
+        os.remove("temp_chunk.wav")
+    if os.path.exists("converted_audio.wav"):
+        os.remove("converted_audio.wav")
+    if os.path.exists(audio_filepath):
+        os.remove(audio_filepath)
+
+    return aligned_chunks
+
+file_transcribe = gr.Interface(
+    fn=transcribe_speech_local,
+    inputs=gr.Audio(sources="upload", type="filepath"),
+    outputs=gr.JSON(label="Transcription with Time Alignment"),
+    allow_flagging="never"
+)
+
+link_transcribe = gr.Interface(
+    fn=transcribe_speech_from_youtube,
+    inputs=gr.Textbox(lines=1, placeholder="Enter YouTube URL here...", label="YouTube URL"),
+    outputs=gr.JSON(label="Transcription with Time Alignment"),
+    allow_flagging="never"
+)
+
+demo = gr.TabbedInterface(
+    [file_transcribe, link_transcribe ],
+    ["Local files(mp3/mp4/wav)", "Links"]
+)
+
+demo.launch(share=True)

requirements.txt

@@ -0,0 +1,6 @@
+transformers
+gradio
+torch
+torchaudio
+pydub
+pytube