Remlemon/yourmt3-vocal-enhanced

A MIDI transcription model focuses on vocal, fine-tuned on the MIR-ST500 Chinese vocal.

Files

• —model.pt - inference weights.
• —config.json - model/audio/task configuration.
• —r47b_infer.py - local vocal stem to MIDI CLI.
• —runtime_deps/, yourmt3_train_src/ - minimal runtime source.

Usage

pip install -r requirements.txt
python r47b_infer.py vocals.wav --output-midi vocals.mid --device cuda

Input should be a separated vocal stem.

Metrics

Metrics are reported on MIR-ST500 vocal transcription splits; this model repo does not include dataset audio or labels.