pip install -r requirements.txt

Note: Before running the scripts, you need to modify the paths in the scripts according to your setup.

Run transcription and technique classification separately.

# Step 1. Audio to MIDI
# Edit AUDIO_DIR, OUTPUT_DIR, and CHECKPOINT_PATH in the script before running
bash scripts/run_HRPT_inference.sh

# Step 2. Recognize playing techniques for each note
# Requires audio directory and corresponding MIDI directory from Step 1
# Edit AUDIO_DIR, MIDI_DIR, OUTPUT_DIR, and checkpoint paths in the script before running
bash scripts/run_HRPT_inference_note_tech.sh

Transcription and technique classification in a single step — no pre-existing MIDI required.

# Audio → MIDI + per-note technique labels (CSV) in one pass
# Edit AUDIO_DIR, OUTPUT_DIR, and checkpoint paths in the script before running
bash scripts/run_infer_technique.sh

• scripts/run_HRPT_inference.sh: Convert audio files to MIDI format

  • Supports .wav, .mp3, .flac formats
  • Can process single file or entire directory
  • Configuration required: Edit the following variables at the top of the script:
    • AUDIO_DIR: Path to audio file or directory
    • OUTPUT_DIR: Output directory for MIDI files
    • CHECKPOINT_PATH: Path to transcription model checkpoint

• scripts/run_HRPT_inference_note_tech.sh: Recognize playing techniques for each note

  • Requires audio directory and corresponding MIDI directory
  • Outputs playing techniques for each note to CSV file
  • Configuration required: Edit the following variables at the top of the script:
    • AUDIO_DIR: Directory containing audio files
    • MIDI_DIR: Directory containing corresponding MIDI files
    • OUTPUT_DIR: Output directory for technique CSV files
    • NOTE_MODEL_CHECKPOINT: Path to note technique model
    • TRANSCRIPTOR_CHECKPOINT: Path to transcriptor model

• scripts/run_infer_technique.sh: End-to-end transcription + technique classification

  • Takes raw audio as input; no pre-existing MIDI needed
  • Outputs per-note technique labels to CSV and a technique-annotated MIDI file
  • Uses the full model (all transcription features, no ablation)
  • Configuration required: Edit the following variables at the top of the script:
    • AUDIO_DIR: Directory containing audio files (.wav, .mp3, .flac)
    • OUTPUT_DIR: Output directory for CSV and MIDI files
    • NOTE_MODEL_CHECKPOINT: Path to note technique model (checkpoints/note_tech_model.pth)
    • TRANSCRIPTOR_CHECKPOINT: Path to transcription model checkpoint