• •Kyutai and Mirelo released MuScriptor, an open-weight transformer model designed to convert multi-instrument audio directly into MIDI format.
• •The model relies on a large-scale proprietary training dataset to handle complex audio polyphony.
• •The release lacks detailed benchmarks comparing transcription accuracy against established proprietary competitors like Ableton's extract tools or Melodyne.

Kyutai has released MuScriptor, an open-weight AI model designed to transcribe multi-instrument audio into editable MIDI data. Unlike traditional spectral analysis tools that often struggle with instrument separation, MuScriptor uses a transformer-based architecture trained on a massive audio dataset to map complex waveforms to musical notation. However, the model remains unproven in professional production environments where precision is paramount, as independent performance benchmarks have yet to be published. Whether this open-weight approach can rival industry-standard proprietary software will depend on its ability to handle nuanced performance data without requiring significant manual correction.