This project provides tools for downloading YouTube videos, performing speaker diarization, and transcribing the audio. It leverages state-of-the-art models for accurate speaker identification and transcription.

• Download audio from YouTube videos
• Perform speaker diarization to identify different speakers in the audio
• Transcribe the audio to text
• Save detailed protocols of speaker segments and their transcriptions

For transcription, we use the Whisper turbo model from OpenAI. Whisper is a general-purpose speech recognition model that is trained on a large dataset of diverse audio.

For downloading audio from YouTube, we use the pytubefix library.

For speaker diarization, we use the pyannote.audio library. pyannote.audio is a toolkit for speaker diarization that provides pre-trained models for speaker identification and segmentation.

To install the required dependencies, run:

uv sync

You need to set the HUGGING_FACE_TOKEN environment variable. You can do this by adding the following line to your shell configuration file (e.g., .bashrc, .zshrc):

export HUGGING_FACE_TOKEN=<your_hugging_face_token>

To use Celery, you need to have Redis installed and running. You can install Redis using Homebrew:

brew install redis

Start the Redis server:

brew services start redis

To monitor Celery tasks, you can use Flower. Install Flower using pip:

uv add flower

To monitor Celery tasks and run Celery workers, follow these steps:

1. Start the Flower server:

   uv run celery -A diarization.celery_task flower --port=5555

   Access the Flower dashboard by navigating to http://localhost:5555 in your web browser.

2. Start the Celery worker:

   uv run celery -A diarization.celery_task worker --loglevel=info -P threads

To convert the audio to WAV format, you need to have ffmpeg installed. You can install ffmpeg using Homebrew:

brew install ffmpeg

To download and process a single YouTube video, run:

uv run diarization.py <YouTube_URL> <output_folder>

To process multiple YouTube URLs from a simple text file (one link per line), run:

uv run diarization.py --file <file_with_urls> <output_folder>

To run Celery tasks for downloading and transcribing, start the Celery worker:

uv run celery -A diarization.celery_task worker --loglevel=info -P threads

Then, you can call the task from your code:

from diarization.celery_task import download_and_transcribe result = download_and_transcribe.delay(<YouTube_URL>, <output_folder>) print(result.get())

To run the Streamlit app for a user-friendly interface, use the following command:

uv run streamlit run streamlit_app.py

The output will include:

• The downloaded audio file in WAV format
• A JSON file with diarization results
• A text file with detailed protocols of speaker segments and their transcriptions

uv run diarization.py https://www.youtube.com/watch?v=example <output_folder>

The diarization results can be used for various applications, including:

• Training large language models (LLMs) with speaker-specific data
• Implementing Retrieval-Augmented Generation (RAG) applications for more accurate and context-aware responses

This project is licensed under the MIT License.