MuScriptor — Music Transcription (Audio → MIDI)

Upload a music recording and get a downloadable MIDI file with transcribed notes. MuScriptor is a ~1.3B parameter multi-instrument automatic music transcription model developed by Mirelo x Kyutai.

Model card · Code · Audio samples

Upload or record audio

Instrument conditioning (optional)

Select instruments present in the audio to improve transcription accuracy. Leave empty for automatic detection.

acoustic_piano electric_piano chromatic_percussion organ acoustic_guitar clean_electric_guitar distorted_electric_guitar acoustic_bass electric_bass violin viola cello contrabass orchestral_harp timpani string_ensemble synth_strings voice orchestra_hit trumpet trombone tuba french_horn brass_section soprano_and_alto_sax tenor_sax baritone_sax oboe english_horn bassoon clarinet flutes synth_lead synth_pad drums

Beam size

1 = Fast (greedy), 4 = High quality (slower).

1 8

Batch size

Parallel GPU chunk processing. Higher = much faster (default 8).

1 16

ZeroGPU Duration limit (seconds)

Time hint requested from ZeroGPU scheduler. Keep at 60-90s when using batch=8.

30 180

Use sampling (stochastic decoding)

If enabled, uses temperature-based sampling instead of greedy/beam decoding.

Temperature

Sampling temperature (only used when sampling is enabled).

0.1 2

Download MIDI file

Examples

·

Built with Gradio logo

·

·