Shazam-inspired music recognition system — FAISS vector search + spectral fingerprinting, with CLAP/MuQ embeddings, FastAPI backend and React frontend.

Song identification combining landmark audio fingerprinting with CNN/AST/MERT embeddings (contrastive InfoNCE), benchmarked on Jamendo with strong noise robustness.

Transfer sonic characteristics from reference sounds to a source audio using Music2Latent embeddings

Universal audio embeddings + tagging + similarity search with Streamlit demo and FastAPI; PANNs/YAMNet, FAISS/Qdrant, Grad-CAM.

Music-emotion alignment via contrastive learning. Pairs frozen MERT audio encoder with BGE text encoder in a shared 512D space for emotion-based music retrieval.

Multimodal extractors for video, image, audio, text & PDF — turn any file into searchable vector embeddings (SigLIP, Gemini, E5, CLAP, ArcFace).

multi-agent AI music production orchestrator. semantic sample search, voice-to-MIDI, Ableton control, stem separation, MIDI generation, MCP server.

Self-hosted semantic music search and audio tagging over a catalog on Backblaze B2. Essentia extracts BPM, key, genre, and mood; CLAP embeddings power find-similar and plain-English text-to-audio search. All local OSS models — B2 is the only credential.