Feature Set Optimisation for Infant Cry Classification

ALBORNOZ, ENRIQUE M.; VIGNOLO, LEANDRO D.; MARTÍNEZ C.

Advances in Artificial Intelligence - IBERAMIA 2018

Springer

Lugar: Cham; Año: 2018; p. 455 - 466

This work deals with the development of features for the automaticclassification of infant cry, considering three categories: neutral,fussing and crying vocalisations. Mel-frequency cepstral coefficients, togetherwith standard functional obtained from these, have long been themost widely used features for all kind of speech-related tasks, includinginfant cry classification. However, recent works have introduced alternativefilter banks leading to performance improvements and increasedrobustness. In this work, the optimisation of a filter bank is proposed forfeature extraction and two other spectrum-based feature sets are compared.The first set of features is obtained through the optimisation offilter banks, by means of an evolutionary algorithm, in order to finda more suitable speech representation for the infant cry classification.Moreover, the classification performance of the optimised representationcombined with other spectral features based on the mean log-spectrumand auditory spectrum is evaluated. The results show that these featuresets are able to improve the performance for the cry classification task.