KeyIT

KeyIT is a macOS musical key detector using CoreML + Accelerate, based on a CoreML port of MusicalKeyCNN.

CLI

Example:

./build/keyit-cli --input /path/to/song.mp3

Options:

• -i, --input <path> audio input (required)
• -m, --model <path> model path (.mlmodelc or .mlpackage)
• --ml-input <name> CoreML input feature name (default spec)
• --ml-output <name> CoreML output feature name (default logits)
• --max-seconds <sec> cap analyzed duration from start (default 480)
• --ml-cpu-only force CPU-only CoreML execution
• --cqt-log-bias-calibration <path> per-bin float list for CQT log-bias calibration
• --dump-features <path> write CQT features as CSV (rows = bins)
• --dump-probs <path> write probabilities as 24 lines
• --bench print benchmark timings
• --verbose verbose diagnostics
• -h, --help usage

CLI output includes:

• top prediction (Class, Camelot, Key, Confidence)
• ambiguity status (Ambiguous, margin)
• alternate class/camelot/key when ambiguous
• top-5 class probabilities

Analysis pipeline

Core path:

1. decode to mono float32 (AVFoundation)
2. resample to target rate (default 44.1kHz)
3. compute log-CQT-like matrix (105 bins by default)
4. trim trailing frames (default 2)
5. harmonic time-median enhancement (enabled by default)
6. windowed CoreML inference (default window/hop 100/100)
7. probability aggregation with frame-energy gating
8. optional section-voted probability blending (enabled by default)
9. ambiguity margin / alternate prediction derivation

Advanced knobs are exposed in keyit::KeyitConfig (include/keyit/keyit.h).

Library API

Public API:

• keyit::estimate_key_from_samples(...)
• keyit::compute_log_cqt_features_from_samples(...)
• keyit::camelot_label(...)
• keyit::key_name_label(...)

See include/keyit/keyit.h for config and result structures (KeyitConfig, KeyEstimate, PipelineTiming).

Credits

KeyIT builds on ideas and model work from MusicalKeyCNN.

Please cite and refer to the original publication for scientific use and further reading:

[1] F. Korzeniowski and G. Widmer. "Genre-Agnostic Key Classification With Convolutional Neural Networks". In: Proceedings of the 19th International Society for Music Information Retrieval Conference (ISMIR) (2018) arXiv

[2] F. Korzeniowski and G. Widmer. "End-to-End Musical Key Estimation Using a Convolutional Neural Network". In: Proceedings of the 25th European Signal Processing Conference (2017) arXiv