README.md

metadata

title: AI Music Attribution
emoji: 🎵
colorFrom: purple
colorTo: blue
sdk: docker
app_port: 7860
pinned: false
hf_oauth_scopes:
  - email

AI Music Attribution

Analyze AI-generated music to identify potential training data sources using MERT embeddings.

Features

• 🎵 Upload audio tracks for analysis
• 🔍 Chunk-level similarity matching with 10K+ training tracks
• 📊 Visual attribution results with waveform display
• 🎼 Support for MP3, WAV, OGG formats
• 🔐 HuggingFace OAuth authentication

How It Works

1. Sign in with your HuggingFace account
2. Upload your audio file
3. Analyze using MERT music embeddings
4. View matched training tracks with similarity scores

Technology

• Frontend: React + Vite + Tailwind
• Backend: Node.js + tRPC
• ML: MERT-v1-95M (768-dim music embeddings)
• Matching: Cosine similarity with dynamic thresholds

Resources

• MERT Model: m-a-p/MERT-v1-95M