A Hierarchical Latent Mixture Model for Polyphonic Music Analysis

Abstract

Polyphonic music transcription is a challenging problem, requiring the identification of a collection of latent pitches which can explain an observed music signal. Many state-of-the-art methods are based on the Non-negative Matrix Factorization (NMF) framework, which itself can be cast as a latent variable model. However, the basic NMF algorithm fails to consider many important aspects of music signals such as lowrank or hierarchical structure and temporal continuity. In this work we propose a probabilistic model to address some of the shortcomings of NMF. Probabilistic Latent Component Analysis (PLCA) provides a probabilistic interpretation of NMF and has been widely applied to problems in audio signal processing. Based on PLCA, we propose an algorithm which represents signals using a collection of low-rank dictionaries built from a base pitch dictionary. This allows each dictionary to specialize to a given chord or interval template which will be used to represent collections of similar frames. Experiments on a standard music transcription data set show that our method can successfully decompose signals into a hierarchical and smooth structure, improving the quality of the transcription.

Improving piano music signal recognition through enhanced frequency domain analysis

“…Waghmare et al [8] conducted a study on the classification and labeling of Indian music, proposed that Mel-frequency cepstral coefficients (MFCCs) can provide timbre information, and demonstrated the effectiveness of this method through experimental analysis. O'Brien et al [9] conducted a study on the transcription of polyphonic music and proposed a probabilistic latent component analysis model. Their experiments demonstrated that this method effectively decomposed the signal into distinct hierarchical smooth structures, resulting in high-quality transcriptions.…”