New open-source Dettato model transcribes audio to MIDI at 20M parameters
A developer released Dettato, an open-source audio-to-MIDI transcription model with 20 million parameters, trained without any copyrighted recordings. On published benchmarks across datasets like Bach10, RWC and Slakh2100, Dettato scores higher than MuScriptor 1.3B and the closed-source Mirelo A2M v1.1 Pro, despite being roughly 65 times smaller than MuScriptor. Its outputs are permitted for commercial use.
GoKawiil's interpretation of the reporting above, not reported fact.
If the benchmark results hold up under independent testing, a much smaller open model outperforming larger and closed competitors could lower the cost and barrier for developers building music transcription tools. The claim of training on no copyrighted recordings may also appeal to users wary of copyright disputes affecting AI training data, though this detail is self-reported by the creator rather than independently verified.
- Dettato is a 20M-parameter open-source audio-to-MIDI model, about 65x smaller than MuScriptor 1.3B
- It reportedly outperforms MuScriptor and closed-source Mirelo A2M v1.1 Pro on published benchmarks
- The model was trained without copyrighted recordings and permits commercial use of its outputs
Source: kobi.music, 2026-11-17
Published there as: “Show HN: I made an audio-to-MIDI model 60x more efficient”
Read the original report → The summary and analysis above are GoKawiil's own, written from reporting by the source above. Facts and quotes belong to the original publisher.