The paper presents an automated system for quantifying non‑linguistic vocal social signals—activity level, engagement, stress, and mirroring—using a two‑level Hidden Markov Model to segment speech and model turn‑taking dynamics. These measures are extracted from the first five minutes of a face‑to‑face salary‑package negotiation between a Middle Manager (MM) and a Vice President (VP) in 46 gender‑matched dyads of MIT Sloan students. The authors demonstrate that the combined voice‑based predictor explains 57% of the variance in objective negotiation outcomes (r=0.57, p=0.001). Status‑specific analyses reveal that for VPs, engagement plus stress accounts for 75% of outcome variance (r=0.75), while for MMs, mirroring alone explains 57% (r=0.57). The same signals also correlate with subjective impressions of influence and conflict avoidance. The work situates these findings within prior research on thin‑slice judgments and underscores the potential of automated vocal signal analysis for real‑time assessment of interpersonal behavior.
Topics
Non‑linguistic Vocal Signals as Predictors of Negotiation Success
Introduces the premise that tone of voice and prosodic style, though often unconscious, can predict behavioral outcomes in negotiations.
Four Quantitative Voice‑Based Social Measures: Activity, Engagement, Stress, Mirroring
Defines each of the four signal dimensions, their theoretical grounding, and how they map onto observable vocal patterns.
Hidden Markov Model Framework for Segmenting Speech and Modeling Turn‑Taking
Details the two‑level HMM process that separates voiced/non‑voiced segments and models participant turn‑taking to compute engagement and other measures.
Design of the Salary‑Package Negotiation Experiment with MIT Sloan Dyads
Describes the participant pool, gender‑matched dyads, negotiation scenario, data collection (voice recordings, objective scores, subjective questionnaires), and duration.
Predictive Findings: Thin‑Slice Voice Signals Correlate with Objective and Subjective Outcomes
Presents the main statistical results: overall predictor r=0.57, status‑specific r values, and correlations with impression and conflict‑avoidance ratings.
Status‑Specific Effects: VP Engagement/Stress vs. MM Mirroring
Explores how high‑status participants (VPs) benefit from engagement and stress signals, whereas lower‑status participants (MMs) benefit from mirroring.
Broader Implications for Automated Social Signal Analysis
Discusses the potential of automated vocal signal measurement for real‑time assessment in various interpersonal contexts beyond negotiation.