YouTube Video Question Answering

This project allows you to perform question answering on YouTube videos by transcribing the video content and using a language model to answer questions based on the transcription.

Features

• Download audio from YouTube videos.
• Transcribe audio to text.
• Split transcriptions into manageable chunks.
• Create embeddings for document chunks.
• Store documents in a vector store for efficient retrieval.
• Answer questions based on the transcribed content.

Installation

To get started with this project, follow these steps:

1. Clone the repository:

   git clone <repository-url>
   cd <repository-directory>

2. Create a virtual environment:

   python3 -m venv venv
   source venv/bin/activate

3. Install the required dependencies:

   pip install -r requirements.txt

4. Set up environment variables:

   Create a .env file in the root directory of the project and add the following environment variables:

   GROQ_API_KEY=<your-groq-api-key>
   HUGGINGFACEHUB_API_TOKEN=<your-huggingfacehub-api-token>

Why Transcribe Audio This Way?

The transcription process involves converting audio to 16kHz mono and splitting it into chunks before transcribing. This approach is taken for several reasons:

• Consistency and Quality: Converting audio to a standard format (16kHz mono) ensures consistent quality and compatibility with the transcription API.
• Manageable Chunks: Splitting the audio into smaller chunks makes the transcription process more manageable and reduces the likelihood of errors or timeouts.

Usage

1. Run the Streamlit app:

   streamlit run rag_app.py

2. Provide the YouTube video link:

   When prompted, enter the YouTube link of the video you want to transcribe and analyze.

3. Ask questions:

   After the transcription process is complete, you can start asking questions based on the transcribed content. Leave the question blank to quit the program.

How It Works

1. Download Audio:

   • The script downloads the audio from the provided YouTube link using pytubefix.

2. Transcribe Audio:

   • The audio is converted to 16kHz mono and split into chunks.
   • Each chunk is transcribed using the Groq API.

3. Process Transcription:

   • The transcription is saved to a file and loaded for further processing.
   • The text is split into chunks for efficient retrieval.

4. Create Embeddings:

   • Embeddings for the document chunks are created using a HuggingFace model.

5. Store Documents:

   • The documents are stored in a vector store for efficient retrieval.

6. Answer Questions:

   • A retrieval chain is set up to answer questions based on the transcribed content using a language model.

Dependencies

• langchain
• langchain[docarray]
• langchain_groq
• langchain_community
• langchain_huggingface
• docarray
• pydantic==1.10.8
• pytubefix
• python-dotenv
• tiktoken
• ruff
• pypdf
• groq
• streamlit