BibTeX

@inProceedings{beccaria-etal-2023-extraction-334169,
	title        = {Extraction and Analysis of Acoustic Features from Italian-Speaking Children with Autism Spectrum Disorder
},
	abstract     = {Background: The persistent difficulties in social interaction and communication that characterize Autism Spectrum Disorder can be accessed by investigating the quality of language. Indeed, these deficits involve the presence of anomalies in speech production and understanding, which find an expression at the acoustic and prosodic levels of linguistic analysis.
Objectives: The main aim of this work is to propose a speech pipeline for the extraction of Italian speech biomarkers typical of ASD by conducting an acoustic and phonological analysis. Moreover, we will highlight the strengths and difficulties of this kind of investigation introducing new topics for further research.
Methods: The poster will present the analysis of a speech corpus of 14 Italian-speaking children with ASD and 14 controls (C). The corpus is demographically balanced (age 6-10, 8;1 ± 1;3. Sex: 3F, 11 M) and homogeneous at the diatopic level (origin: Prato, Pistoia, Florence).
First, we extracted the acoustic features by using eGeMAPS (openSMILE; Eyben et al., 2015), specifically ideated for the study of impaired speech. Then, we implemented the Mann-Whitney U-test to select the features with the most statistically significant distance in the production of the two groups.
Secondly, we conducted a parallel extraction regarding the pitch (F0 mean and standard deviation). We propose this additional analysis because pitch varies according to some demographic traits of the speaker (sex, age, height) and the literature presents opposite trends. For this task, we used Praat to have more flexibility in the manipulation of the extraction. We set the F0 range between 70 and 400 Hz (Patel et al., 2020).
Finally, we conducted a comparison between the results of the two methods excluding female participants to verify if the trend of pitch changes when the participants are not mixed.
Results: Table 1 shows the features selected between the ones extracted. They are related to prosody, quality of voice, loudness, and spectral distribution.
Jitter, shimmer and HNR are usually investigated together to describe the emotional prosody and the quality of voice. The same trend found on our corpus is recorded in previous studies on languages other than Italian (Bone et al. 2015; Kissine & Geelhand 2019). Moreover, spectral flux is usually investigated together with shimmer and jitter to describe speech impairments (Haider et al., 2019). Nevertheless, if we consider the studies related to autistic speech, there are few that describe this feature because of the different methodologies used during the extraction.
Finally, the values of pitch extracted by eGeMAPS and Praat show the same trend. It is higher in ASD than in controls, both if we considered the corpus mixed and the one with only the male speakers. However, the pitch does not show a statistically significant difference between the two groups (Table 2).
Conclusions: These results, although preliminary, seem to confirm the presence of phonetic alterations of speech associated with the disorder. Further studies could improve the accuracy of the pipeline proposed by doing a qualitative analysis of the results and considering other linguistic and paralinguistic domains (e.g., morphological, pragmatic, and gestural analysis).
},
	booktitle    = {The 22nd International Society for Autism Research (INSAR)},
	author       = {Beccaria, Federica and Gagliardi, Gloria and Kokkinakis, Dimitrios},
	year         = {2023},
}