Basic cuts revisited: Temporal segmentation of speech into phone-like units with statistical learning at a pre-linguistic level

Okko RäsänenAalto University, Department of Signal Processing and Acoustics

Abstract

Considerable effort has been put to understand how infants may utilize statistical regularities of speech in early word segmentation. Some studies suggest that infants are able to discover word boundaries at the points of high unpredictability across subsequent linguistic units such as phonemes or syllables. Meanwhile, the possible role of the statistical regularities in the temporal organization of the speech at a pre-linguistic acoustic level has not been widely addressed. The current work examines how the short-term temporal predictability of the acoustic speech signal correlates with linguistically motivated phone-, syllable-, and word-level units. The results indicate that the points of low predictability correlate mainly with the boundaries between phone-like segments. This suggests that the same statistical learning mechanisms hypothesized to operate at the word level can also aid in temporal organization of the speech stream into phone-like temporal segments before knowing the phonemic or syllabic units of the language.

Files

Basic cuts revisited: Temporal segmentation of speech into phone-like units with statistical learning at a pre-linguistic level (732 KB)



Back to Table of Contents