Speech-based Diagnosis of Autism Spectrum Condition by Generative Adversarial Network Representations

International audience; Machine learning paradigms based on child vocalisations show great promise as an objective marker of developmental disorders such as Autism. In conventional detection systems, hand-craaed acoustic features are usually fed into a discriminative classiier (e. g., Support Vector Machines); however it is well known that the accuracy and robustness of such a system is limited by the size of the associated training data. is paper explores, for the rst time, the use of feature representations learnt using a deep Genera-tive Adversarial Network (GAN) for classifying children's speech aaected by developmental disorders. A comparative evaluation of our proposed system with diierent acoustic feature sets is performed on the Child Pathological and Emotional Speech database. Key experimental results presented demonstrate that GAN based methods exhibit competitive performance with the conventional paradigms in terms of the unweighted average recall metric.