Kyutai Releases MuScriptor: Open-Source Decoder-Only Transformer for Multi-Instrument Audio-to-MIDI
Decision Brief
MuScriptor, from Kyutai and Mirelo, is an open-weight decoder-only Transformer for transcribing multi-instrument audio to MIDI. It was trained on 170k real recordings and 1.45M synthetic MIDI pieces using a three-stage pipeline. The article compares it to YourMT3+ and covers instrument conditioning, setup, and interactive demos. For researchers and developers working on music transcription or audio processing, MuScriptor provides freely deployable open weights without commercial API dependency, lowering the barrier for polyphonic transcription. The combined real-synthetic training strategy boosts generalization, making it ideal for teams handling complex mixes in real-world scenarios.
Sources
- MarkTechPost
Fast research-paper and ML tooling summaries, useful for infra and agent updates.
- MarkTechPost
留言
登入后即可留言,和其他 builder 交换实测心得。
还没有留言,抢头香。