跳到正文
hackernews100(kobi.music)· hidude562·· 6 天前精選品質評分79

Dettato 推出音訊轉 MIDI 模型,引數量為 20M

Show HN: I made an audio-to-MIDI model 60x more efficient

重點導讀

把音樂錄音轉成 MIDI,模型大小和訓練資料都會影響使用方式。

  • 作者推出開源模型 Dettato,表示它有 20M 引數,在文中列出的公開基準上,分數高於 MuScriptor 1.3B,也略高於 Mirelo A2M v1.1 Pro。
  • 訓練資料使用合成樂器錄音與 URMP 資料集中的 Creative Commons 獨奏樂器錄音,沒有使用受著作權保護的錄音。
  • 作者也提醒,模型不擅長正確分開樂器,而且輸出的是節拍對齊的演奏 MIDI,不是可直接演奏的樂譜。

AI 伴隨問答

自備 Key 深度分析

針對這篇報導有疑問?可直接點選快捷問題或向 AI 追問背景脈絡與技術細節:

推薦理由

Dettato 用 20M 引數在文中列出的公開基準取得較高分數,也交代了不使用受著作權保護錄音訓練的做法。

正文 · 原文

16 November 2026

A low-poly pink bird, drawn ten times in a loop, listening

we're introducing Dettato, a SOTA audio to MIDI transcription model that significantly beats MuScriptor 1.3B - the next best open model - and scores a bit higher than mirelo's closed source model 'Mirelo A2M v1.1 Pro' on their published benchmarks. all at just 20m parameters - roughly 65x smaller than muscriptor. Additionally, in training the model, we trained on absolutely no copyrighted recordings. the model is open source and you can use its outputs for commercial purposes.

the composite, and the measures it is the mean of, for each system
systemcompositeonsetsframesoffsetsdrumsparameters
dettato60.870.879.140.952.520M
Mirelo A2M v1.1 Pro56.167.175.740.141.4not published
MuScriptor47.457.071.231.430.11.3B
YourMT3+45.758.263.930.530.046M
sets used: PHENICX, RWC classical, RWC royalty-free

F1 per set: the mean of onset, frame and offset

F1 per set: the mean of onset, frame and offset
setdettatomirelo A2M v1.1 promuscriptor 1.3Byourmt3+
Bach1091.3not reported54.757.9
Dagstuhl ChoirSet60.8not reported35.528.0
PHENICX-Anechoic64.860.954.444.8
RWC Popular54.0not reported44.336.1
RWC Classical64.465.858.462.4
RWC Genre48.7not reported42.635.6
RWC Jazz55.7not reported52.047.1
RWC Royalty-Free61.656.246.845.4
Slakh2100 (synthetic)68.4not reported51.1not reported

disclaimer with the benchmarks: it's difficult to gauge the true accuracy of the model for your use-case without trying it yourself. one thing my model performs poor on for instance, is splitting instruments correctly. In addition, this just spits out performance MIDI with a beat aligned. If you're looking for something that you can actually play(e.g sheet music), i'd reccommend songscription or klangio - both are excellent tools by great companies that push the envelope of what's possible in this field forward.

0:00 / 0:00

sousa, the stars and stripes forever: “the president's own” united states marine band, from the complete marches of john philip sousa, vol. 3. a us government recording, public domain. download the MIDI

upload your own example(or pick a demo below)

the driving goal of kobimusic: to help musicians, not replace them. a part of that goal means following copyright law. we trained dettato on a mix of synthetic instrument recordings and CC solo instruments mixed in using the URMP dataset. the rest of the instruments morph together ~20 soundfonts, using the 'soundfont babies' technique i published. no copyrighted recordings were used.

I began working on dettato about 10 months ago, but scrapped what I had in about April. The reason why was that I treated AMT as a data problem - but in reality, the bottleneck is architecture. There are a lot of parellels that you can draw for OMR (first model i made) and AMT. both traditionally were tackled early on using algorithms. later, in both fields they plugged in giant LLM components. These models would naturally become overfit to synthetic data, since globally the problems of AMT and OMR become different than if you treat it locally. But, for both you can reframe it as a local problem. For OMR, using a detector model. For AMT, a pitch-lane model.

overall, i hope both open and closed weight music retrieval companies can learn from this. The constraint of not using copyrighted recordings doesn't hold you back, but actually encourages innovation, and it's likely you'll end up with a better model in the end.

來源: hackernews100 (kobi.music)
開啟原文

讀者討論

理性交流 · 即時討論