Unsupervised classification of music genre using hidden Markov model

Abstract

Music genre classification can he of great utility to musical database management. Most of current classification methods are supervised and tend to he based on contrived taxonomies. However, due to the ambiguities and inconsistencies in the chosen taxonomies, these methods are not applicable for much larger database. In this paper, we proposed an unsupervised clustering method based on a given measure of similarity which can he provided by Hidden Markov Models. In addition, in order to better characterize music content, a novel segmentation scheme is proposed based on music intrinsic rhythmic structure analysis and features are extracted based on these segments. The performance of this feature segmentation scheme performs better than the traditional fixed-length method according to experimental results. Our preliminary results also suggest that proposed method is comparable to supervised classification method.

Discrimination Between Native and Non-Native Speech Using Visual Features Only

“…The idea of using HMMs to directly model time-sequential information as a whole, rather than separately in sub-units, is not new. For instance, continuous HMMs have been successfully employed in this fashion to categorise music clips into music genres [56]. Also, Schuller et al [57] approach emotion recognition by feeding acoustic low-level descriptors, such as adjusted pitch and energy contours along with their first-and second-order derivatives, into emotion-specific continuous HMMs.…”

Music Genre Classification: A Semi-supervised Approach

“…Music files are clustered basing on an objective function to dynamically build a taxonomy depending on the clustering outcome. Shao et al [5] used Agglomerative Hierarchical Clustering.…”

Dirichlet Process HMM Mixture Models with Application to Music Analysis

“…The term L(q) can be obtained by substituting (13) and (14) into (12). The optimization of the lower bound L(q) is realized by taking functional derivatives with respect to each of the q(·) distributions [3].…”

“…Since "the brain dynamically links a multitude of short events which cannot always be separated" [2], temporal cues are critical and contain information that should not be ignored. Therefore, music is treated as time-series data and hidden Markov models (HMMs), which can accurately represent the statistics of sequential data [8], have been introduced to model the overall music in [2] [9] and more recently for music genre classification [10] [12].…”