app.py

import gradio as gr
import ffmpeg
import os
import pytube
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
import torch
import numpy as np
import wave

# Load Wav2Vec 2.0 processor and model for transcription
processor = Wav2Vec2Processor.from_pretrained("facebook/wav2vec2-base-960h")
model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-base-960h")

# Function to extract audio from video
def extract_audio(video):
    if "youtube" in video:
        yt = pytube.YouTube(video)
        video_stream = yt.streams.filter(progressive=True, file_extension="mp4").first()
        video_stream.download(filename="temp_video.mp4")
        video_path = "temp_video.mp4"
    else:
        video_path = video.name
        video.save(video_path)
    
    # Extract audio using ffmpeg
    audio_path = "temp_audio.wav"
    ffmpeg.input(video_path).output(audio_path).run()
    
    return audio_path

# Function to transcribe audio using Wav2Vec 2.0
def transcribe_audio(audio_path):
    with wave.open(audio_path, 'rb') as f:
        frames = f.readframes(f.getnframes())
        waveform = np.frombuffer(frames, dtype=np.int16)
        input_values = processor(waveform, return_tensors="pt", padding=True).input_values
        logits = model(input_values).logits
        predicted_ids = torch.argmax(logits, dim=-1)
        transcription = processor.decode(predicted_ids[0])
    return transcription

# Summarize the transcript using a Hugging Face summarization model
from transformers import pipeline
summarizer = pipeline("summarization")

def summarize_text(text):
    summary = summarizer(text, max_length=200, min_length=50, do_sample=False)
    return summary[0]['summary_text']

# Gradio interface function
def gradio_interface(video, question=None):
    audio_path = extract_audio(video)
    transcript = transcribe_audio(audio_path)
    summary = summarize_text(transcript)
    
    answer = None
    if question:
        answer = "This is where a Q&A feature would go."  # Placeholder for Q&A
    
    return summary, answer if question else summary

# Gradio Interface
iface = gr.Interface(
    fn=gradio_interface,
    inputs=[
        gr.Video(label="Upload Your Video or YouTube Link"),
        gr.Textbox(label="Ask a Question (Optional)", lines=2)
    ],
    outputs=[
        gr.Textbox(label="Summary"),
        gr.Textbox(label="Answer (If Question Asked)")
    ],
    title="VideoGPT: AI Video Summarizer + Q&A",
    description="Upload a video or paste a YouTube link to extract the audio, get a summary, and ask questions about the video content."
)

# Launch the interface
iface.launch()