Integrating Changes of Register Into Automatic Intonation Analysis

Céline de Looze and Daniel Hirst, Laboratoire Parole et Langage, CNRS et Université de Provence

While current tools for the automatic analysis and modeling of intonation are satisfactory for laboratory or isolated sentences, they appear insufficient for the study of longer stretches of authentic speech, which are in general marked by systematic changes of register. This study shows that implementing automatically detected register changes significantly improves the accuracy of the automatic coding of intonation patterns with the INTSINT algorithm. This implies, upstream, to define a reliable measurement of register and a way to detect its changes automatically.