Piano2notes — Mod

This modifies the local machine learning parameters. The official model prioritizes speed; the mod adjusts the "temperature" and "beam search width" of the AI.

  • MIDI pipeline:
  • Post-processing heuristics:

    • As of late 2025, the original Piano2Notes company has started using "model fingerprinting." Every output MIDI file now contains a cryptographic hash of your user ID. Consequently, mods that merely unlock the API now produce traceable files. The community has pivoted toward local-only mods that run entirely offline using open-source models (like the piano-transcription-inference fork).

    The next generation of the mod (v3) reportedly includes: piano2notes mod

  • Edit values and restart Minecraft for changes to take effect.

    • Example mapping table (conceptual)

    | Configuration | Note F1 (full) | |---------------|----------------| | Full system | 0.83 | | – HMM tracking | 0.76 | | – onset head (frame only) | 0.71 | | – data augmentation | 0.79 | This modifies the local machine learning parameters

    The HMM provides the largest gain (+7 points) by removing spurious notes.

    We use a Convolutional Recurrent Neural Network: MIDI pipeline:

    Loss function:
    L = L_onset + L_frame
    where both are binary cross-entropy.

    Training uses MAESTRO v3.0 (1286 hours of aligned MIDI + audio). Data augmentation: pitch shift (±2 semitones), time stretch (0.9×–1.1×), and background noise.