Acoustic-Prosodic and Physiological Response to Stressful Interactions in Children with Autism Spectrum Disorder

Daniel Bone, Julia Mertens, Emily Zane, Sungbok Lee, Shrikanth S. Narayanan, Ruth B. GrossmanInterspeech 2017 · DOI

Introduction

Social anxiety affects approximately 40% of individuals with Autism Spectrum Disorder (ASD), creating significant barriers to effective communication and daily functioning. A major challenge in clinical management is that individuals with ASD often struggle to introspectively identify or verbally communicate their internal stress levels. While "atypical prosody"—unusual rhythm, pitch, or intonation—is a recognized hallmark of ASD, traditional research often examines speech in isolation. This study addresses this gap by employing a "Behavioral Signal Processing" approach, using engineering tools to create objective, scalable measures of social stress by analyzing dynamic changes in speech and physiology (heart rate) across varying social demands.

Methodology

The researchers utilized a subset of the USC CARE Corpus, involving 41 children (17 with ASD, 24 neurotypical/TD) with an average age of 13.5 years. To induce a gradient of social stress, a three-tier experimental protocol was implemented:

  1. Low Stress: A casual one-minute conversation about YouTube clips.
  2. Medium Stress: A formal 20-minute interview regarding personal topics like hobbies and school.
  3. High Stress: A 5-minute oral presentation delivered to a panel of "judges" (simulated via video) where participants believed they were being evaluated against their peers.

Data collection included video-recorded interactions and heart rate monitoring. The team extracted several feature categories: acoustic-prosodic (pitch, intensity, high-frequency energy), speaking rate, prosodic coordination (synchronization of pitch, intensity, and duration), and physiology (mean heart rate). Support Vector Machines (SVM) were then employed to classify both the stress level of the task and the diagnostic group.

Key Insights and Results

The study's most significant finding was the power of relative features. Analyzing how an individual's speech changed from a low-stress baseline to a high-stress task proved far more predictive than absolute measurements. This suggests that atypicality in ASD is best characterized as a specific pattern of deviation from one's own baseline.

Table of feature trends across task stress levels and ASD diagnosis.

As shown in the feature analysis, increased stress led to reduced coordination between pitch, intensity, and duration across both groups, likely indicating a breakdown in motor control under pressure. ASD-specific markers included higher prosodic variability but a flatter pitch slope within syllables, often perceived as "monotone" or "awkward."

Changes in fundamental frequency (f0) and heart rate (HR) across stress conditions for ASD and TD groups.

The classification results were robust:

  • Stress Classification: Acoustic features identified task stress levels with a 70% Unweighted Average Recall (UAR).
  • ASD Diagnosis: The system achieved its peak performance—87% UAR—when using relative acoustic changes between low and high-stress scenarios to distinguish ASD from TD.
  • Task Sensitivity: Diagnostic accuracy was significantly higher in the "High Stress" task (84%) compared to the "Low Stress" task (65%), suggesting that social pressure accentuates the behavioral markers of autism.

Conclusion and Future Directions

This research demonstrates that multimodal analysis of speech and physiology can provide objective markers for social stress in children with ASD. While the sample size (N=41) is a limitation requiring further validation, the results highlight the importance of baseline-relative measurements. Future work will focus on larger cohorts and more sensitive physiological measures, such as Heart Rate Variability (HRV), to better map the relationship between internal physiological states and external vocal behaviors.

Summary updated

BibTeX
@inproceedings{bone2017acoustic,
  author = {Daniel Bone and Julia Mertens and Emily Zane and Sungbok Lee and Shrikanth S. Narayanan and Ruth B. Grossman},
  title = {{Acoustic-Prosodic and Physiological Response to Stressful Interactions in Children with Autism Spectrum Disorder}},
  booktitle = {Interspeech},
  year = {2017},
  doi = {10.21437/interspeech.2017-179},
}
Figure