The Prosogram model for pitch stylization and its applications in intonation transcription

This chapter first describes pitch stylization based on a tonal perception model, as implemented in the Prosogram tool. The key element in pitch perception is the segmentation of speech into syllable-sized portions, as a result of spectral change and sonority variation. The pitch variations of these speech portions are subject to perceptual thresholds, resulting in simplified shapes synchronized with the segmental layer. After an outline of tonal perception, we describe the stylization algorithm, its validation through resynthesis, and its application to prosodic feature extraction. The second part of the chapter describes two applications of this stylization to intonation transcription. The first type transcribes pitch variation into discrete levels and movements, for each syllable in the speech signal. The second type takes into account prosodic structure and identifies intonation contours as sequences of pitch events anchored in prosodic structure. The proposed bottom-up analysis starts from the acoustic signal and computes successive representations of increasing degree of abstraction. Implemented as a computational model, it illustrates a generic, integrated and interdisciplinary approach to intonation analysis, which may be applied to many languages in a semi-automatic way.status: accepte