bluenevus's picture
Update app.py
17ca647 verified
raw
history blame
4.58 kB
import gradio as gr
import torchaudio
import torchaudio.transforms as T
from transformers import pipeline
import requests
from pydub import AudioSegment
from pydub.silence import split_on_silence
import io
import os
from bs4 import BeautifulSoup
import re
import numpy as np
from moviepy.editor import VideoFileClip
import soundfile as sf
from spellchecker import SpellChecker
import tempfile
# Load the transcription model
transcription_pipeline = pipeline("automatic-speech-recognition", model="facebook/wav2vec2-base-960h")
spell = SpellChecker()
def download_audio_from_url(url):
try:
if "share" in url:
print("Processing shareable link...")
response = requests.get(url)
soup = BeautifulSoup(response.content, 'html.parser')
video_tag = soup.find('video')
if video_tag and 'src' in video_tag.attrs:
video_url = video_tag['src']
print(f"Extracted video URL: {video_url}")
else:
raise ValueError("Direct video URL not found in the shareable link.")
else:
video_url = url
print(f"Downloading video from URL: {video_url}")
response = requests.get(video_url)
audio_bytes = response.content
print(f"Successfully downloaded {len(audio_bytes)} bytes of data")
return audio_bytes
except Exception as e:
print(f"Error in download_audio_from_url: {str(e)}")
raise
def correct_spelling(text):
words = text.split()
corrected_words = [spell.correction(word) or word for word in words]
return ' '.join(corrected_words)
def format_transcript(transcript):
sentences = transcript.split('.')
formatted_transcript = []
current_speaker = None
for sentence in sentences:
if ':' in sentence:
speaker, content = sentence.split(':', 1)
if speaker != current_speaker:
formatted_transcript.append(f"\n\n{speaker.strip()}:{content.strip()}.")
current_speaker = speaker
else:
formatted_transcript.append(f"{content.strip()}.")
else:
formatted_transcript.append(sentence.strip() + '.')
return ' '.join(formatted_transcript)
def transcribe_audio(video_bytes):
try:
with open("temp_video.mp4", "wb") as f:
f.write(video_bytes)
video = VideoFileClip("temp_video.mp4")
audio = video.audio
audio.write_audiofile("temp_audio.wav", fps=16000, nbytes=2, codec='pcm_s16le')
audio_data, sample_rate = sf.read("temp_audio.wav")
if len(audio_data.shape) > 1:
audio_data = audio_data.mean(axis=1)
audio_data = audio_data.astype(np.float32) / np.max(np.abs(audio_data))
result = transcription_pipeline(audio_data)
transcript = result['text']
transcript = correct_spelling(transcript)
transcript = format_transcript(transcript)
os.remove("temp_video.mp4")
os.remove("temp_audio.wav")
return transcript
except Exception as e:
print(f"Error in transcribe_audio: {str(e)}")
raise
def transcribe_video(url):
try:
print(f"Attempting to download audio from URL: {url}")
audio_bytes = download_audio_from_url(url)
print(f"Successfully downloaded {len(audio_bytes)} bytes of audio data")
print("Starting audio transcription...")
transcript = transcribe_audio(audio_bytes)
print("Transcription completed successfully")
return transcript
except Exception as e:
error_message = f"An error occurred: {str(e)}"
print(error_message)
return error_message
def download_transcript(transcript):
with tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt') as temp_file:
temp_file.write(transcript)
temp_file_path = temp_file.name
return temp_file_path
# Create the Gradio interface
with gr.Blocks(title="Video Transcription") as demo:
gr.Markdown("# Video Transcription")
video_url = gr.Textbox(label="Video URL")
transcribe_button = gr.Button("Transcribe")
transcript_output = gr.Textbox(label="Transcript", lines=20)
download_button = gr.Button("Download Transcript")
download_link = gr.File(label="Download Transcript")
transcribe_button.click(fn=transcribe_video, inputs=video_url, outputs=transcript_output)
download_button.click(fn=download_transcript, inputs=transcript_output, outputs=download_link)
demo.launch()