Acoustic Measurements of Prosodic Information in Toddlers with Autism Spectrum Disorders

Background: Disordered prosody (modulation of speech to clarify meaning) has long been considered a hallmark of Autism Spectrum Disorders (ASD; Kanner 1943), and atypical intonation has recently been recognized as an important diagnostic ASD marker (Gotham 2007). Prosodic characterization of children with ASD is an under-researched area, particularly for very young and pre-verbal children, although studies suggest vocal atypicality may represent an early appearing symptom of ASD (Sheinkopf et al 2000). Studies have traditionally evaluated prosody by subjective perceptual analysis (McCann & Peppe 2003). Few studies have utilized objective computerized acoustic voice analysis to measure prosodic elements in children with ASD (e.g., Diehl 2009, Green & Tobin 2009, Oller 2010). No acoustic studies to date have focused on toddlers with ASD, when early diagnosis is often still a challenge.  

Objectives: To measure and quantify parameters of intonation, volume and vocal quality using computerized acoustic voice analysis, in toddlers with ASD in comparison to non-ASD controls.

Methods: : Data were derived from the ongoing longitudinal prospective Canadian “Infant Sibling Study” (see Zwaigenbaum et al. 2012). Participants were 23 younger siblings of probands with ASD, followed throughout infancy and independently diagnosed with ASD at age 3 years using the Autism Diagnostic Observation Schedule (ADOS) and Autism Diagnostic Interview – Revised (ADI-R), and an age-matched comparison group (n=20) with no family history of ASD and confirmed as not having ASD at age 3 years. All the toddlers’ vocalizations were extracted from video-recorded ADOS assessments conducted at 18 months of age. Vocalizations were converted to audio, de-identified, and digitalized. Non-overlapped segments were acoustically analyzed, using the VoiceSauce MATLAB toolbox optimized for children's speech. A total of 26 acoustic parameters were extracted, related to intonation (pitch), maturity of speech (first formant frequencies and amplitudes), volume (energy) and measures of vocal quality such as voice breathiness, harshness/creakiness (harmonics, spectral tilt and cepstral peak prominence). Variations on these parameters between groups have been widely used to identify speech disorders. ADOS intonation scores were also compared between groups.  

Results: We compared mean, standard deviation (std) and range for the acoustic parameters and found significant between-group differences for 11 parameters, including volume (mean, std, and range), pitch (range) and nine other parameters of vocal quality, using family-wise error rate corrections (p-value range: 0.0001-0.045). Five parameters, the most robust of which was volume, remained significant when subjected to formal Bonferroni correction (critical p = .05/26 =.0019). For volume, the mean and range values were more than twice as large for the ASD group (p=0.0005). We found no between-group differences for length of ADOS assessment or total duration of analyzed vocalizations, but found a significant group difference for clinician-rated intonation score (p=.002); in all cases, values were higher for the ASD group.  

Conclusions: We identified and characterized, using objective acoustic measures, early-appearing vocal differences in toddlers later diagnosed with ASD. These findings may contribute to efforts aimed at early detection. Future plans include evaluation at earlier ages (9 and 12 months) and associations between acoustic variables and clinical ratings of intonation.