Automatic Classification of Infant- and Adult-directed Speech from Acoustic Speech Signals

Abstract

Paralinguistic speech processing (PSP) is a field of audio processing where the focus of the analysis is beyond the literal message. One potential task in the area of PSP is the classifica-tion of samples into infant-directed speech (IDS) and adult-directed speech (ADS). In the pre-sent study, a system which classifies samples into IDS/ADS as accurately as possible was examined by experimenting with different classifiers used in other fields of PSP, and by testing different sets of manually defined features. The classification results showed that the best set of features was a set which included all speech-relevant features extracted in this study except spectrogram. Additionally, support vector machines (SVMs) performed the best of the individ-ual classifiers used in the study, while an ensemble classifier outperformed all individual clas-sifiers. These results were in line with the previous IDS/ADS classification studies in the field.

Keywords: classification, speaking style, infant-directed speech, adult-directed speech, motherese, speech processing, acoustic signal