The Temporal Structure of the Autistic Voice: A Cross-Linguistic Investigation

Background:  Individuals with autism spectrum disorder (ASD) tend to have atypical modulation of speech and voice, often described as awkward, monotone, or sing-songy [1-3]. These anomalies may constitute one of the most robust and fast signals of social communication deficits in this population [4, 5].  However, it has proven difficult to determine a consistent set of acoustic features that can account for these perceived differences.  Using Recurrence Quantification analysis of acoustic features, Fusaroli et al. [6] demonstrated a high efficacy of identifying voice patterns characteristic of adult Danish speakers with Asperger’s syndrome. 

Objectives:  We systematically quantify and explore speech patterns in children with and without autism across two languages: Danish and American English.  We employ traditional and non-linear techniques measuring the structure (regularity and complexity) of speech behavior (i.e. fundamental frequency, use of pauses, speech rate). Our aims are (1) to achieve a more fine-grained understanding of the speech patterns in children with ASD, and (2) to employ the results in a supervised machine-learning process to determine whether acoustic features can be used to predict diagnostic status within and across languages.

Methods:  Our analysis was based on previously-acquired repeated narratives (TOMAL-2 [7]) in Danish, and a story retelling task [1] in American English). We tested 25 Danish and 25 US children diagnosed with ASD as well as 25 Danish and 16 US matched controls. Age range was 8-13 years with no significant difference between language groups. Transcripts were time-coded, and pitch (F0), speech-pause sequences and speech rate were automatically extracted. For each prosodic feature we calculated recurrence quantification measures, that is, the number, duration and structure of repeated patterns[8]. The results were employed to train a linear discriminant function algorithm to classify the descriptions as belonging either to the ASD or the control group, using 1000 iterations of 10-fold cross-validation (to test the generalizability of the accuracy) and variational Bayesian mixed-effects inferences (to compensate for biases in sample sizes). Algorithms were trained on Danish data only, American English data only and the combined group, to investigate the presence of cross-linguistic features of prosodic patterns in ASD.

Results:  Voice recordings within each language group were classified with balanced accuracy, sensitivity and specificity all > 77% (p<.000001), The cross-linguistic corpus was classified with balanced accuracy, sensitivity and specificity all >71% (p<.000001). Voices of individuals with ASD can be characterized as more regular (that is, with patterns regularly repeated) in their pitch and pause structure and more irregular in speechrate.

Conclusions:  Non-linear recurrence analyses techniques suggest that there are quantifiable acoustic features in speech production of children with ASD that distinguish them from typically developing speakers, even across linguistic and cultural boundaries.