Music and speech in early development: automatic analysis and classification of prosodic features from two Portuguese variants

Abstract

In the present study, we aim to capture rhythmicand melodic patterning in speech and singing directed to infants. We addressthis issue by exploring the acoustic features that best predict differentclassification problems. We built a database composed by infant-directed speechfrom two Portuguese variants (European vs Brazilian Portuguese) and infant-directedsinging from the two cultures, comprising 977 tokens. Machine learning experimentswere conducted in order to automatically discriminate between language variantsfor speech, vocal songs and between interaction contexts. Descriptors related with rhythm exhibited strongpredictive ability for both speech and singing language variants’discrimination tasks, presenting different rhythmic patterning for each variant.Moreover, common features could be used by a classifier to discriminate speechand singing tasks, indicating that the processing of speech and singing might sharethe analysis of the same properties of the stimuli. With respect todiscrimination between different interaction contexts, pitch-relateddescriptors showed better performance. Therefore, we conclude that prosodiccues present in the surrounding sonic environment of an infant are sources ofrich information not only to make distinction between different communicativecontexts through melodic cues, but also to provide specific cues about therhythmic identity of their mother tongue. These prosodic differences may leadto further research on their influence in infant’s development of musicalrepresentations.