Abstract
Atypical speech prosody is a primary characteristic of autism spectrum disorders (ASD), yet it is often excluded from diagnostic instrument algorithms due to poor subjective reliability. Robust, objective prosodic cues can enhance our understanding of those aspects which are atypical in autism. In this work, we connect objective signal-derived descriptors of prosody to subjective perceptions of prosodic awkwardness. Subjectively, more awkward speech is less expressive (more monotone) and more often has perceived awkward rate/rhythm, volume, and intonation. We also find expressivity can be quantified through objective intonation variability features, and that speaking rate and rhythm cues are highly predictive of perceived awkwardness. Acoustic-prosodic features are also able to significantly differentiate subjects with ASD from typically developing (TD) subjects in a classification task, emphasizing the potential of automated methods for diagnostic efficiency and clarity.
Introduction
Prosody—the rhythm, stress, and intonation of speech—serves as a vital social cue in human interaction. While "atypical prosody" is a hallmark of Autism Spectrum Disorder (ASD), it remains notoriously difficult to measure objectively. Clinical diagnostic tools often omit prosodic assessment because clinicians frequently struggle to reach a consensus on what specifically constitutes "awkward" or "atypical" speech. This research, presented at INTERSPEECH 2015, seeks to bridge the gap between subjective human perception and objective signal processing by identifying the specific acoustic "fingerprints" of awkward prosody in adolescents with ASD.
Methodology
The study involved 69 adolescents (43 with ASD and 26 typically developing) who participated in a story-retelling task. The research team employed a dual approach to analyze the speech samples:
- Perceptual Ratings: Fifteen naive human raters evaluated 322 utterances on Likert scales for overall "awkwardness," "expressivity," and specific components including rate/rhythm, volume, and intonation.
- Acoustic Feature Extraction: Thirty-seven objective features were computed, covering rate and rhythm (e.g., articulation rate, pausing), voice quality (e.g., jitter, shimmer), and intonation (e.g., pitch and intensity slopes).
Notably, the researchers introduced exemplar-based features. They established a "gold standard" prosodic template from highly-rated speakers and measured the deviation of each subject's speech from this template using Spearman’s correlation.
Results and Key Insights
Human Perception and Agreement
Human raters demonstrated a high level of agreement regarding "expressivity" ($\rho = 0.70$) and moderate agreement for "overall awkwardness" ($\rho = 0.57$). However, raters were less consistent when identifying the specific cause of the awkwardness (e.g., volume vs. intonation), suggesting that while humans are sensitive to prosodic oddness, the underlying drivers are complex.
Drivers of Awkwardness
The analysis revealed that Rate and Rhythm features are the primary predictors of perceived awkwardness. Specifically, increased pausing and high variability in vowel duration were strongly associated with higher awkwardness ratings. Speech perceived as awkward was also strongly correlated with being "less expressive" or monotone, a quality that could be objectively quantified through pitch and intensity variance.
Diagnostic Classification
Using only Rate and Rhythm features, a logistic regression model successfully distinguished between ASD and typically developing (TD) subjects with a 69% Unweighted Average Recall (UAR). Furthermore, the automated "Feature Fusion" model predicted human awkwardness ratings with a correlation of 0.56, nearly matching the level of agreement between human raters.
Conclusion
This work demonstrates that automated acoustic analysis can effectively quantify the subjective perception of "awkwardness" in ASD speech. Timing and rhythm emerge as the most critical markers of prosodic atypicality. These findings suggest that objective, signal-derived features have significant potential to enhance diagnostic efficiency and provide clearer targets for clinical intervention. Future research will explore these markers in spontaneous, conversational contexts to further validate their diagnostic utility.
Summary updated