RAG Over Audio Files

A new app demonstrates Retrieval-Augmented Generation (RAG) over audio files using AssemblyAI for transcription and DeepSeek-R1 as the language model. The process involves transcribing audio, storing it in a Qdrant vector database, querying the database for context, and generating responses. AssemblyAI's latest Universal-2 model shows significant improvements in speech recognition, outperforming competitors across multiple languages. The app, built with a simple Streamlit interface, allows users to upload and chat with audio files directly.