Paralinguistic analysis of children's speech in natural environments

Abstract

Paralinguistic cues are the non-phonemic aspects of human speech that convey information about the affective state of the speaker. In children's speech, these events are also important markers for the detection of early developmental disorders. Detecting these events in hours of audio data would be beneficial for clinicians to analyze the social behaviors of children. The chapter focuses on the use of spectral and prosodic baseline acoustic features to classify instances of children's laughter and fussing/crying while interacting with their caregivers in naturalistic settings. In conjunction with baseline features, long-term intensity-based features, that capture the periodic structure of laughter, enable in detecting instances of laughter to a reasonably high degree of accuracy in a variety of classification tasks.