summarizer.py

import yt_dlp
from faster_whisper import WhisperModel
from ollama import Client
import os
import browser_cookie3
import tempfile

def get_youtube_cookies():
    """
    Attempt to get YouTube cookies from multiple browsers.
    Returns a cookie file path or None if no cookies found.
    """
    browsers = [
        (browser_cookie3.chrome, 'chrome'),
        (browser_cookie3.firefox, 'firefox'),
        (browser_cookie3.edge, 'edge'),
        (browser_cookie3.safari, 'safari')
    ]
    
    # Create a temporary file for cookies
    cookie_file = tempfile.NamedTemporaryFile(delete=False, suffix='.txt')
    
    for browser_func, browser_name in browsers:
        try:
            cookies = browser_func(domain_name='.youtube.com')
            with open(cookie_file.name, 'w') as f:
                for cookie in cookies:
                    f.write(f'{cookie.domain}\tTRUE\t{cookie.path}\t'
                           f'{"TRUE" if cookie.secure else "FALSE"}\t{cookie.expires}\t'
                           f'{cookie.name}\t{cookie.value}\n')
            return cookie_file.name
        except:
            continue
    
    return None

def download_audio(url):
    cookie_file = get_youtube_cookies()
    
    ydl_opts = {
        'format': 'bestaudio/best',
        'postprocessors': [{
            'key': 'FFmpegExtractAudio',
            'preferredcodec': 'wav',
            'preferredquality': '192',
        }],
        'outtmpl': 'audio.%(ext)s',
        'quiet': True,
        'no_warnings': True
    }
    
    # Add cookies if available
    if cookie_file:
        ydl_opts['cookiefile'] = cookie_file
    
    try:
        with yt_dlp.YoutubeDL(ydl_opts) as ydl:
            ydl.download([url])
            
        # Clean up cookie file
        if cookie_file and os.path.exists(cookie_file):
            os.unlink(cookie_file)
            
        return 'audio.wav'
    except Exception as e:
        if cookie_file and os.path.exists(cookie_file):
            os.unlink(cookie_file)
        raise Exception(f"Error downloading audio: {str(e)}")

def transcribe_audio(audio_file):
    # Explicitly specify CPU device and compute type
    model = WhisperModel("base", device="cpu", compute_type="int8")
    segments, _ = model.transcribe(audio_file)
    return " ".join([segment.text for segment in segments])

ollama_client = Client()

def generate_notes_and_summary_stream(transcript):
    prompt = f"""
    Based on the following transcript of a video, create:
    1. A set of concise, informative notes
    2. A brief summary of the main points

    Transcript:
    {transcript}

    Notes and Summary:
    """
    
    stream = ollama_client.generate(model='llama3.1:latest', prompt=prompt, stream=True)
    for chunk in stream:
        yield chunk['response']

def process_video(url):
    try:
        audio_file = download_audio(url)
        transcript = transcribe_audio(audio_file)
        return generate_notes_and_summary_stream(transcript)
    except Exception as e:
        raise Exception(f"An error occurred: {str(e)}")