Describe a sound you can play
SEED is a generative synthesizer controlled through language.
Describe the sound you want. AI turns it into a synthesizer keyboard you can define, control and play.
Describe
Describe the texture, dynamics and character you imagine.
Define
Organize complex parameters into meaningful, playable controls.
Play
Make music with the keyboard, knobs, velocity and your own judgment.
Concept rendering.
TimbreCLAP
“A warm pad with a slow attack.” A source name alone cannot describe onset, texture and evolution. TimbreCLAP adapts CLAP with timbre-focused captions so text search lands closer to the sound itself.
- 64.2%
- Core-1K text–audio retrieval R@10, vs 14.3% for CLAP-music
- 0.173
- Mean correlation with published timbre ratings, vs 0.020 for CLAP-music
- 708,606
- instrument notes and synth presets captioned for training
ISMIR 2026 Late-Breaking Demo. Retrieval results are single-seed and not tested for statistical significance.
Coming in 2027
Sign in or register to get SEED launch updates.