How I Built a Prompt-to-Music AI Agent & Browser-Based Karaoke Separator with React & ONNX
The article details the development of two AI-powered music tools: a prompt-to-music agent that generates audio from text descriptions, and a browser-based karaoke separator that isolates vocals from instrumentals. Both are built with React for the frontend and ONNX (Open Neural Network Exchange) for running machine learning models directly in the browser. The karaoke separator uses a pre-trained model to split audio into vocal and accompaniment tracks, while the music generation agent leverages a text-to-music model. The author emphasizes the use of ONNX Runtime Web for efficient client-side inference, avoiding server costs and latency. The project is part of a retro-themed challenge, with CSS styling mimicking Windows 95 and hacker aesthetics. The tools are designed to be accessible and interactive, allowing users to experiment with AI music processing without specialized hardware or cloud services.
Enables client-side AI music generation and separation, reducing server dependency and latency.