As group size increases, individuals modify their vocal features to signal cooperation while remaining recognizable.

Introduction: Cooperation is essential to humans and manifests in speech through acoustic convergence, in which interlocutors' voices become more similar. Yet convergence can be limited when signaling individuality is more important than aligning with others. In such contexts, speakers may adopt non-accommodative strategies to preserve their vocal identity and recognizability. Drawing on animal communication research-which shows that species in larger social groups exhibit greater vocal individuality-we test how group size (3 and 5 interactants) shapes the balance between remaining identifiable while still cooperating to support communication. Methods: In an interactive online game, players collaborated on a shared task while trying to recognize one another by voice, with the player recognized best receiving a reward. To assess how group size influences accommodation under these dual demands, we analyzed the speech of three players who participated in both game sessions. Acoustic features relevant to voice identity and amenable to accommodation-harmonicity, jitter, F0, formant dispersion, and duration-were extracted and reduced through Principal Component Analysis to two dimensions accounting for 52.9% of the variance. Following the identification of group size effects on these components, we assessed whether inter-speaker acoustic differences increased, decreased, or remained stable across conditions. Additionally, within-speaker variability was examined as a function of group size to determine whether observed changes in the five-player condition were driven by all speakers or only a subset. Results: In larger groups, accommodation was selective: players modulated their acoustic features, maintaining some while converging on others. No significant differences as a function of group size were observed for PC1, interpreted as reflecting maintenance and primarily associated with voice quality measures (harmonicity and jitter). In contrast, significant differences emerged for PC2, largely driven by F0 standard deviation and duration. These changes indicated reduced inter-speaker differences in the larger group, consistent with convergence. In the larger group setting, convergence was observed when two speakers showed mutual alignment and also shifted toward the speech patterns of a third participant, who remained comparatively stable in their acoustic behavior. Discussion: The pattern of selective and asymmetrical convergence indicates that speakers strategically balance the goals of cooperation and individuality, suggesting that recognizability demands also shape accommodation.