Automated extraction of speech and turn-taking parameters in autism allows for diagnostic classification using a multivariable prediction model

Abstract

Autism spectrum disorder (ASD) is diagnosed on the basis of speech and communication differences, amongst other symptoms. Since conversations are essential for building connections with others, it is important to understand the exact nature of differences between autistic and non-autistic verbal behaviour and evaluate the potential of these differences for diagnostics. In this study, we recorded dyadic conversations and used automated extraction of speech and interactional turn-taking features of 54 non-autistic and 26 autistic participants. The extracted speech and turn-taking parameters showed high potential as a diagnostic marker. A linear support vector machine was able to predict the dyad type with 76.2% balanced accuracy (sensitivity: 73.8%, specificity: 78.6%), suggesting that digitally assisted diagnostics could significantly enhance the current clinical diagnostic process due to their objectivity and scalability. In group comparisons on the individual and dyadic level, we found that autistic interaction partners talked slower and in a more monotonous manner than non-autistic interaction partners and that mixed dyads consisting of an autistic and a non-autistic participant had increased periods of silence, and the intensity, i.e. loudness, of their speech was more synchronous.

Can Natural Speech Prosody Distinguish Autism Spectrum Disorders? A Meta-Analysis

“…Therefore, features with remarkable contributions to distinguish ASD from TD groups were selected with tools like correlation analysis [53,67], principal component analysis, factor analysis [18,62], ElasticNet [63][64][65], and Geneva Minimalistic Acoustic Parameter Set (GeMAPS) [68]. Data selection was further classified by tools such as native Bayed (NB) [42], support vector machines (SVMs) [5,20,41,60,66,68], probabilistic neural networks (PNNs) [19], speech-related vocal islands (SVIs) [62], or random forests [67]. Since machine learning was not merely to find a model explaining the current data but to create a model that generalizing to new data [69].…”

“…Given that only six studies were included in the meta-analysis, further moderator analysis and model construction were neglected. Forest plot for the meta-analysis of pitch variability differences [3,16,29,34,51,52,54,56,58,60].…”

“…Figure 6. Forest plot for the meta-analysis of pitch variability differences[3,16,29,34,51,52,54,56,58,60].…”

Neural processing of social reciprocity in autism

“…Such insights into the differentiation of interpersonal synchrony have sparked investigation into its relevance for the diagnostic classification of autism (Georgescu et al, 2019;Koehler et al, 2021Koehler et al, , 2022Plank et al, 2023). However, the mechanistic root underlying altered interpersonal synchrony, as a function of atypical reciprocity, in autism has not yet been explored.…”

“…For example, dyads including an autistic individual (or dyads of two individuals with autism) tend to show less synchronous behavior than dyads including two non-autistic individuals (Georgescu et al, 2020). Such insights into the differentiation of interpersonal synchrony have sparked investigation into its relevance for the diagnostic classification of autism (Georgescu et al, 2019; Koehler et al, 2021, 2022; Koehler & Falter-Wagner, 2023; Plank et al, 2023). However, the mechanistic root underlying altered interpersonal synchrony, as a function of atypical reciprocity, in autism has not yet been explored.…”