An acoustic-perceptual study of the telephone speaking style with implications for speaker verification in Brazilian Portuguese

Author(s)

Renata Regina Passetti

• Document type: Doctoral Thesis
• Press: Campinas, SP.
• Institution: Universidade Estadual de Campinas (UNICAMP). Instituto de Estudos da Linguagem
• Defense date: 2018-06-06

Examining board members

Plinio Almeida Barbosa; Pablo Faria; Ana Carolina Constantini; Sandra Madureira; Pablo Arantes

• Advisor: Plinio Almeida Barbosa

Abstract

This research presents a phonetic-experimental analysis of the acoustic parameters that characterize the telephone speaking style and also investigates the speakers¿ ability of discriminating a speaking style representative of mobile phone interactions from a non-mediated speaking style representative of face-to-face interactions. A main corpus was compiled in an attempt to study the production of the telephone speaking style, within that a sample was selected to analyze the perceptual dimension of this speaking style. The main corpus, named "EstiloTel", consisted of semi-spontaneous speech samples which were recorded in face-to-face and telephone conditions and obtained in noisy and silent environments. The goals of this stage were to investigate if the acoustic modifications of telephone speaking style could be influenced by environmental noises and also if they could be distinct between sexes. The acoustic analysis techniques were divided considering global and local parameters, besides the durational analysis of silent and filled pauses, inter-silent pauses, and stress groups. In general, the telephone speaking style is linked to an increase of median values of fundamental frequency for both sexes. However, male speakers modify a greater amount of acoustic parameters during a telephone interaction than women. Among males, the telephone speaking style is characterized by a significant change of fundamental frequency statistical descriptors and temporal parameters, namely distribution of normalized duration peaks, inter-silent pauses duration, and stress groups duration. Moreover, changes related to most of these parameters are enhanced by environmental noises. Excluding inter-silent pauses duration, the remaining telephone condition significant parameters tended to be higher when measured in noisy environments. On the other hand, inter-silent pauses duration rate is lower in telephone condition which signals shorter speech excerpts and a higher number of silent pauses in this condition. Therefore, male speech in telephone interactions has higher and more variable values of fundamental frequency, from a durational point of view is produced with more salience, and has a higher amount of silent pauses. The study of the perceptual dimension has shown that telephone speaking style is auditorly hard to identify. On the other hand, the analysis of listeners¿ sensibility across stimuli distinct grouping has shown a higher accuracy in the telephone speaking style identification when the interactions of both conditions takes place in the same environment. The correlation between the listeners' choices in the perceptual experiment and the acoustical parameters was investigated by non-parametric regression analysis. The results have shown that approximately one third of the telephone speaking style identification could be explained by the combination of pitch and loudness-related parameters (AU)

• FAPESP's process: 15/12174-9 - An acoustic-perceptual study of the telephone speaking style with implications for speaker verification in Brazilian Portuguese
• Grantee: Renata Regina Passetti
• Support Opportunities: Scholarships in Brazil - Doctorate