Detecção automática de fronteiras prosódicas entre unidades entonacionais

Abstract

Speech is segmented into intonational units marked by prosodic boundaries. This work aims both to investigate the phonetic-acoustic parameters that guide the production and perception of prosodic boundaries and to develop models for automatic detection of prosodic boundaries in spontaneous speech. Two samples of male spontaneous speech excerpts were segmented into intonational units by two groups of trained annotators. The boundaries perceived by the annotators were annotated as either terminal (TB) or non-terminal (NTB). A script was used to extract phonetic-acoustic parameters along the speech signal. The extracted parameters comprise measures of: 1) Speech rate and rhythm; 2) Normalized duration; 3) Fundamental frequency; 4) Intensity; 5) Silent pause. A training of models composed by multiple parameters designed to the automatic identification of boundaries marked by the annotators was developed. The Linear Discriminant Analysis algorithm was used and positions at which at least 50% of the annotators indicated a boundary of the same type were considered as boundary. The automatic terminal boundary detection model shows a convergence of 80% in relation to terminal boundaries noticed by annotators in sample I. For non-terminal boundaries, three statistical classification models were obtained. Together, the three models show a convergence of 98% in relation to nonterminal boundaries noticed by annotators in sample I. The models were validated later in sample II. The results of the validation indicate that the performance of the TB model is 74% and that of the NTB model is 88% in sample II.