Using a coupled-oscillator model of speech rhythm to estimate rhythmic variability in two Brazilian Portuguese varieties (CE and SP)

Abstract

This paper presents preliminary results of a semi-automatic methodology to extract three parameters of a dynamic model of speech rhythm. The model attempts to analyze the production of rhythm as a system of coupled oscillators which represent syllabicity and phrase stress as levels of temporal organization. The estimated parameters are the syllabic oscillator entrainment rate (alpha), the syllabic oscillator decay rate (beta), and the coupling strength between the oscillators (w0). The methodology involves finding the combination that minimizes the distance between natural duration contours and simulated contours generated using several combinations of the parameters. The distance between natural and model-generated contours was measured in two ways by comparing: (1) plain or overt syllable to syllable duration and (2) relative change along both contours.We applied this methodology to read speech produced by five speakers of the state of Ceará (CE) and eight speakers of the state of São Paulo (SP). Mean w0 and alpha values are compatible with the view that Brazilian Portuguese is a mixed-rhythm language. Results from two bayesian hierarchical regression models do not suggest a difference between SP and CE speakers, but indicate a difference between the two methods, with the relative change method generating lower alpha values and higher w0 values, and the reverse for the plain duration method.

Keywords

• Prosody
• Speech rhythm
• Brazilian Portuguese