Etude des variations individuelles des caractéristiques de la parole et de la voix

Study of individual variations in speech and voice characteristics

• Auteur / Autrice : Estelle De kermel chardenon
• Direction : Christine Meunier, Cécile Fougeron
• Type : Projet de thèse
• Discipline(s) : Sciences du langage
• Date : Inscription en doctorat le 01/10/2020
• Etablissement(s) : Aix-Marseille
• Ecole(s) doctorale(s) : Ecole Doctorale Cognition, Langage et Education
• Partenaire(s) de recherche : Laboratoire : LPL - Laboratoire Parole et Langage

Mots clés

Mots clés libres

Caractéristiques du locuteur

Variation intra-locuteur

Variation inter-locuteurs

Paramètres temporels

Paramètres mélodiques

Speaker characteristics

Intra speaker variation

Inter speakers variation

Temporals parameters

Melodics parameters

Résumé

L'objectif principal de ce projet est de mieux comprendre la variation intra-individuelle des caractéristiques de la voix et de la parole. Nous regarderons l'évolution de la voix et de la parole d'un même individu à travers plusieurs corpus avec des durées et des fréquences d'enregistrement différentes. Pour ce faire, nous analyserons des mesures de modulation, de gestion ou encore d'acoustique des sons (mesures rythmiques, mesures mélodiques) afin d'identifier les indices stables et variables dans la production des locuteurs au travers des enregistrements. Nous observerons ces mesures avec des facteurs pouvant amener à des variations telles que l'interaction, l'aspect temporel ou le style de parole.

Several temporal and melodic parameters have been identified as idiosyncratic (i.e., unique to each speaker) in various studies, reflecting individual anatomy in speech signals. Although analyzing these idiosyncratic parameters might suggest reliable speaker identification in voice recordings, an individual's voice and speech fluctuate constantly and are highly sensitive to variations, even within a single day. Many studies have observed variations in temporal parameters (articulatory rate, modulation of articulatory rate, pause management) and melodic parameters (mean fundamental frequency and variation in fundamental frequency) depending on the speaker's gender, age, quality of life, and emotional state. Other external factors also influence these idiosyncratic parameters: conversational speech, where a speaker may adapt their parameters to the interlocutor; the time span over which speakers are recorded, leading to variations in recordings spaced years apart or during lengthy recordings that can cause vocal fatigue; and individual differences, with speakers standing out in terms of speech rate or voice pitch. To date, numerous studies highlight inter-speaker variations in these parameters, but there is still a lack of knowledge regarding intra-speaker variations, especially in spontaneous speech recordings, which prevents us from fully understanding vocal behavior. The goal of this study is to examine both inter- and intra-speaker variations in these idiosyncratic parameters, alongside the factors that can cause variations in voice and speech. We will investigate these variations using two corpora: the Corpus Interactional Data (CID), which contains 1 hour of spontaneous speech recordings in interaction from 8 dyads, and the Parole AdulTe A TRavers les Ages (PATATRA) corpus, a longitudinal dataset with annual recordings of controlled and spontaneous speech for 11 speakers. Our results show that each speaker is unique in their variations. Some speakers exhibit high variability in certain parameters while remaining very stable in others. They may also show significant differences from other speakers while maintaining stability in their own intra-speaker variations. Furthermore, although some speakers stand out in a specific parameter, this only applies to a minority of them. Inter-speaker variations are numerous across all parameters and are particularly noticeable among speakers with marked differences, such as fast versus slow speech rates. Temporal parameter variations are prominent for the speakers in PATATRA, showing significant differences between annual recordings, which is not the case for CID speakers during an hour-long recording. Melodic parameter variations are widespread in both corpora, showing considerable variations even within the same speaker. We observe many temporal and melodic parameter variations related to speech style when comparing spontaneous speech to controlled speech in PATATRA. These variations are not visible when comparing the two speech styles (narrative vs. non-narrative parts) in CID. Finally, convergence effects are observed in some dyads, although not systematically.