The Relationship Between F0 Synchrony and Speech Convergence in Dyadic Interaction

Abstract

Speech accommodation happens when two people engage in verbal conversation. In this paper two types of accommodation are investigated -one dependent on cognitive, physiological, functional and social constraints (Convergence), the other dependent on linguistic and paralinguistic factors (Synchrony). Convergence refers to the situation when two speakers' speech characteristics move towards a common point. Synchrony happens if speakers' prosodic features become correlated over time. Here we analyze relations between the two phenomena at the single word level. Although calculation of Synchrony is fairly straightforward, measuring Convergence is even more problematic as proved by a long history of debates on how to define it. In this paper we consider Convergence as an emergent behavior and investigate it by developing a robust and automatic method based on Gaussian Mixture Model (GMM). Our results show that high Synchrony of F0 between two speakers leads to greater amount of Convergence. This provides robust support for the idea that Synchrony and Convergence are interrelated processes, particularly in female participants.

The neural oscillatory markers of phonetic convergence during verbal interaction

“…Several studies have focused on subjective evaluations (Pardo, 2013), whereas others have used a variety of objective acoustic measures (Goldinger, 1998). Phonetic convergence was computed using a robust and automatic speaker identification technique applied to the full acoustic spectrum, thus reducing the number of a priori hypotheses about which acoustic feature shows alignment (Mukherjee et al, 2017). One key novelty of our study is that we implemented a quantitative method to extract phonetic convergence from a game-like task, allowing an engaging, yet relatively constrained, phonetic interaction.…”

“…One key novelty of our study is that we implemented a quantitative method to extract phonetic convergence from a game-like task, allowing an engaging, yet relatively constrained, phonetic interaction. Phonetic convergence was computed using a robust and automatic speaker identification technique applied to the full acoustic spectrum, thus reducing the number of a priori hypotheses about which acoustic feature shows alignment (Mukherjee et al, 2017). This method was designed to specifically evaluate cooperative speech behavior.…”

“…We used the Verbal Domino Task (VDT, Mukherjee, D'Ausilio, Nguyen, Fadiga, & Badino, 2017;Mukherjee et al, 2018; adapted version from Bailly & Lelong, 2010) with English words. The VDT is an interactive, collaborative task jointly performed by two participants who pronounce a sequence of disyllabic words with an alternation between participants across words.…”

Analyzing Vocal Tract Movements During Speech Accommodation

“…Part of the complexity is probably due to its dependency on contextual, social and linguistic factors. Previously, our group devised a robust method to extract phonetic convergence in a game-like speech turn-taking task, by using a speaker verification technique [8] [9].…”

“…For this purpose, we evaluated articulatory dynamics underlying speech accommodation. We asked pairs of participants to engage in an interactive speech task [9] while dual-EMA was recorded. Our first aim was to apply the same technique used in [9] to verify its robustness on a different data-set.…”