Semantic Music Tagger and Analyzer
Built a scalable CLAP + TTMR++ hybrid pipeline for holistic music analysis, semantic tagging, caption generation, and similarity search using FMA datasets. Combined multi-vector joint FAISS embeddings with LLMs to generate human-quality descriptions and rich metadata from raw MP3 uploads. Supports text to music recommendation and artist discovery as the barebones of a recommendation algorithm. Built in 72 hours.
Python FastAPICLAPTTMR++FAISSNextJSSentenceTransformersLLaMA2LLMScriptingDemucsFMA Dataset
