RefMatch analyzes your audio and instantly suggests reference tracks that match your mix — by how it actually sounds, not just metadata.
Feed any audio file — WAV, MP3, FLAC, OGG, AIFF, M4A. Works with full mixes, stems, or works in progress.
CLAP neural embeddings capture how your track sounds. DSP extracts 43 technical features — loudness, spectral balance, rhythm, harmony.
Hybrid scoring (60% perceptual + 40% technical) ranks the best matches with explanations of why each track fits.
512-dimensional perceptual model understands how your track sounds — not just its frequency stats. Two tracks that "feel" the same score high.
MFCCs, spectral features, loudness (LUFS), dynamics, tempo, key estimation — the technical characteristics that matter for mixing.
Focus on what matters: match on low-end, loudness, brightness, rhythm, or harmony independently. Find a track with the bass you want.
No cloud, no API calls, no uploads. Your audio stays on your machine. Everything runs locally.
Ships with a CC-licensed database across 25 genres. Ready to use out of the box — or bring your own library.
MIT licensed. The engine is free forever. Build on it, extend it, integrate it into your workflow.
| RefMatch | Spotify Similar | Manual Search | ChatGPT | |
|---|---|---|---|---|
| Analyzes actual audio | Yes | No (collaborative filtering) | Your ears only | No (text only) |
| Mix-level matching | Loudness, spectral, dynamics | Vibe only | If you know what to look for | Guesses from metadata |
| Dimension targeting | Low-end, brightness, rhythm... | No | No | No |
| Works offline | Yes | No | Yes | No |
| Time to match | < 10 seconds | Instant | 10-30 minutes | ~30 seconds |
| Open source | MIT | No | N/A | No |
Be the first to know when the desktop app and VST plugin launch.