Does personality sound the same in stressful and neutral conversations?

Most automatic personality perception (APP) systems treat personality as a static trait independent of context. But psychological research has long acknowledged an interactionist view where both traits and situations shape behavioral expression. This study (2025) provides empirical evidence that perceived personality from conversational speech differs significantly across interaction types.

Key findings: (1) Perceived personalities differ significantly between a neutral interview and a stressful client interaction. (2) In neutral contexts, loudness, sound level, and spectral flux correlate with perceived extraversion, agreeableness, conscientiousness, and openness. In stressful contexts, these same features instead correlate with neuroticism. (3) Handcrafted acoustic features and non-verbal features outperform neural speaker embeddings for personality inference. (4) Stressful interactions are specifically more predictive of neuroticism, consistent with psychological theory that neuroticism manifests under pressure.

The implication for AI persona design: a persona that appears consistent in neutral interactions may fracture under stress. Conversely, an AI system inferring user personality from speech must account for situational context — the same vocal features mean different things in different interaction types. This challenges the design assumption behind systems like Why do open language models converge on one personality type?, where personality is measured in a single context (questionnaire response).

The finding that handcrafted features outperform learned embeddings also suggests that personality perception relies on interpretable paralinguistic cues rather than holistic speaker representations — meaning personality is expressed through specific, measurable behaviors, not through an abstract "speaker style."