✦
Summary
Mirelo Audio-to-MIDI is a music transcription model and product that converts full-mix audio into editable MIDI tracks. Users can upload WAV or MP3 recordings and receive instrument-separated note data for voice, drums, bass, keys, synths, and other detected parts, then review or export the result inside Mirelo Studio.
Key Features
Transcribes full-mix songs into multi-instrument MIDI rather than handling only solo instruments.
Detects and labels instrument parts such as voice, drums, bass, keys, and synth leads.
Outputs separate editable MIDI tracks on a piano-roll interface.
Exports tracks for use in DAWs, notation tools, and music learning workflows.
Represents transcription as a sequence of pitch, timing, and instrument tokens.
Uses synthetic pretraining, real-audio fine-tuning, and post-training for improved accuracy.
Provides open inference code and model weights for researchers.
Includes a free Studio product path for testing the improved production model.
About Mirelo Audio-to-MIDI
The underlying MuScriptor model frames multi-instrument transcription as language modeling over a piano-roll token sequence. It conditions on mel-spectrogram chunks and autoregressively predicts pitch, timing, and instrument tokens, with training that combines large synthetic MIDI pretraining, real audio fine-tuning, and reinforcement-style post-training on curated high-quality transcriptions.
The tool is useful for musicians, producers, composers, educators, and researchers who need to recover musical structure from dense recordings rather than isolated stems. Mirelo and Kyutai open-sourced the research model and inference code while also offering a free product experience in Mirelo Studio with an improved production version.