Unsupervised Discovery of Temporal Structure in Music

Abstract

Abstract-We describe a data-driven algorithm for automatically identifying repeated patterns in music which analyzes a feature matrix using shift-invariant probabilistic latent component analysis. We utilize sparsity constraints to automatically identify the number of patterns and their lengths, parameters that would normally need to be fixed in advance, as well as to control the structure of the decomposition. The proposed analysis is applied to beat-synchronous chromagrams in order to concurrently extract recurrent harmonic motifs and their locations within a song. We demonstrate how the analysis can be used to accurately identify riffs in popular music and explore the relationship between the derived parameters and a song's underlying metrical structure. Finally, we show how this analysis can be used for long-term music structure segmentation, resulting in an algorithm that is competitive with other state-of-the-art segmentation algorithms based on hidden Markov models and self similarity matrices.

Convex non-negative matrix factorization for automatic music structure identification

“…That dataset is composed of 176 songs and is traditionally used to evaluate such segmentation task [10,4,9,12]. We also evaluate against the Internet Archive part of the more recent SALAMI dataset [20], which contains 253 freely available songs.…”

“…We leave an exhaustive exploration of the parameters for future work due to the limitation of space, while still showing that we can obtain good results with this set of arguments. The results of the algorithm are compared against two other techniques that use matrix factorization for music segmentation: SI-PLCA [12] and a variant of our algorithm that uses classic NMF instead of C-NMF. The parameters used for SI-PLCA are the ones proposed for MIREX (see source code 4 ).…”

“…Finally, other methods combine both tasks into one sole algorithm, e.g. using Hidden Markov Models (HMM) [10,11], a probabilistic version of convolutive NMF [12], or k-means clustering [13].…”

Speech Intelligibility Enhancement Using Convolutive Non-negative Matrix Factorization with Noise Prior

“…Certain computational theories such as computational auditory scene analysis (CASA) also rely on such representations [3,8]. As an emerging approach for data analysis, NMF has been successfully used in such applications as music transcription, blind source separation and et al [9][10][11][12].…”

Music self-similarity modeling using augmented nonnegative matrix factorization of block and stripe patterns

“…In [5], NMF is applied to a specific score matrix combining harmonic and timbral information. Moreover, sparse, convolutive NMF has been applied to spectrogram data in order to locate recurrent harmonic motifs as well as to infer high-level structure [6].…”