CASM improves music beat tracking by adapting to local context
CASM: Context-Aware Semi-Markov Post-Processor for Beat Tracking
Sound
Summary
Music beat tracking involves detecting when beats happen in a song, but picking beats directly from signals can be noisy or miss important beats. The authors developed CASM, a new method that adapts to local evidence in the music rather than relying on fixed tempo rules. CASM helps produce smoother and more accurate beat sequences without needing to retrain the main models or adjust settings for different datasets. This makes beat tracking better at handling tricky cases like beats that come at different speeds or ambiguous rhythms.
What this means in practice
- •For music software developers: Integrate CASM into beat tracking modules to improve rhythm extraction accuracy without retraining core neural models.
- •For dance app designers: Use CASM to enhance beat detection in apps that synchronize dance moves to music with varying tempos.
Authors
Zhanhong He, Hanyu Meng, Yaolong Ju
Abstract
In music beat tracking, model-predicted activations must be decoded into a discrete, musically coherent event sequence. Direct peak picking closely follows local evidence but can retain spurious peaks or miss weak beats. Dynamic Bayesian networks (DBNs), a widely used structured post-processor, improve sequence consistency under predefined global tempo, meter, and transition constraints, but their behavior can depend strongly on these settings. We introduce CASM, a context-aware semi-Markov decoder that instead conditions its temporal constraint on local activation evidence. CASM also accounts for ambiguity among competing periodic interpretations, including half- and double-tempo alternatives. Deterministic safeguards prevent implausible outputs and preserve beat-downbeat consistency. Applied to fixed activations from three neural beat trackers (BeatThis, MSCNN, and TCN), CASM improves temporal continuity while preserving event-level F1 across the GTZAN and SMC datasets, without backbone retraining or dataset-specific retuning. Further analysis shows that CASM is less sensitive than the DBN baseline to the composition of the calibration data.