Unique Acoustic Characteristics of Children with Autism and Their Caregivers: A Comparison with Language Delayed and Typically Developing Counterparts

Background:  Our previous research demonstrated the convenience of collecting naturalistic audio data for autism research using wearable recorders. The developed algorithm for automatic data analysis has demonstrated reliability and validity. Naturalistic daylong recordings capture the characteristic behaviors of children, their caregivers, and the interactions between them, providing an ecological means to study the entire natural environment of children with autism. The efficiency of this methodology facilitated data collection for samples of children with autism (ASD), children with language delay (LD) and typically developing children (TD). These data provide a unique opportunity to study both children and caregivers, and to examine acoustic characteristics that are unique to ASD children or their caregivers.

Objectives:  Previous research reports higher f0 and f0 variation for ASD children, based on data from laboratory settings and comparison with TD children only. One objective of this study is to examine f0 using naturalistic data in large quantity and to compare to both TD and LD children. Unique acoustic characteristics for ASD children are studied. Similarly, we examine the acoustic patterns of caregivers to determine whether variations observed in the children hold for adults.

Methods:  Daylong audio recordings were collected using wearable LENA recorders. The automated algorithm detected key-child, adults and other environment sounds. Human voice was further processed via phone recognition algorithms into four sound categories: consonant-like, vowel-like, non-speech-like and pause. We focus on key-child sounds and female adult sounds which are immediately adjacent to key-child and can thus be considered an approximation to caregiver’s child-directed voice. Acoustic phonetic properties of f0 as well as duration, dB-level and spectrum-entropy for each sound category are studied and compared.

Results:  

The dataset comprises 71 ASD children (228-recordings), 49 LD children (333-recordings) and 106 TD children (802-recordings). To some extent, higher f0 and f0 variation for ASD children are confirmed. However, LD children show similar patterns, reducing the uniqueness of this feature to ASD. As well, both ASD and LD children produce longer duration and higher duration variation than TD children. Interestingly, the dB-level of vowel-like sounds and the spectrum-entropy of unvoiced-consonant-like sounds are unique for ASD children compared with both LD and TD children. Caregiver’s child-directed voice usually exhibits longer duration, higher dB and f0 for vowel-like sounds which are characteristics of motherese. These characteristics are further exaggerated for caregivers of ASD children. Non-speech-like sounds, measured with spectrum-entropy, are also unique for the caregivers of ASD children. 

ASD-versus-TD;                ASD-versus-LD;               LD-versus-TD;             Feature

t(175)=5.9, p=1.6e-08;  t(118)=4.8, p=3.9e-06;  t(153)=0.03, p=0.98;  Child-Vowel-dB

t(175)=5.2, p=5.2e-07;  t(118)=3.6, p=5.5e-04;  t(153)=0.74, p=0.46;  Child-unvoiced-consonant-Spectrum-entropy

t(175)=4.6, p=7.7e-06;  t(118)=3.1, p=2.3e-03;  t(153)=0.48, p=0.63;  Caregiver-Vowel-Duration

t(175)=7.1, p=2.9e-11;  t(118)=5.1, p=1.1e-06;  t(153)=0.93, p=0.35;  Caregiver-Vowel-dB

t(175)=3.6, p=3.5e-04;  t(118)=2.8, p=6.7e-03;  t(153)=0.20, p=0.85;  Caregiver-Vowel-f0

t(175)=6.2, p=3.2e-09;  t(118)=4.7, p=6.0e-06;  t(153)=0.71, p=0.48;  Caregiver-Non-speech-like-sound-Spectrum-entropy

Conclusions:  This study demonstrates an ecological way of studying both ASD children and their caregivers in natural home environments using audio recordings. Unique acoustic characteristics for both ASD children and their caregivers are found when compared with LD and TD counterparts. We discuss possible reciprocal effects child vocalization could have on caregiver’s vocal output.