Fast bayesian nmf algorithms enforcing harmonicity and temporal continuity in polyphonic music transcription

This article presents theoretical and experimental results about constrained non-negative matrix factorization (NMF) in a Bayesian framework, enforcing both spectral harmonicity and temporal continuity. We exhibit fast multiplicative update rules to perform the decomposition, which are then applied to perform polyphonic piano music transcription. This approach is shown to outperform other standard NMF-based transcription systems, providing a meaningful mid-level representation of the data.