ISCA Archive Interspeech 2016
ISCA Archive Interspeech 2016

Does She Speak RTT? Towards an Earlier Identification of Rett Syndrome Through Intelligent Pre-Linguistic Vocalisation Analysis

Florian B. Pokorny, Peter B. Marschik, Christa Einspieler, Björn Schuller

For many years, an apparently normal early development has been regarded as a main characteristic of Rett syndrome (RTT), a severe progressive neurodevelopmental disorder almost exclusively affecting girls/females. The speech-language domain represents a key domain for the clinical diagnosis of RTT, which usually happens around three years of age. Recent studies have built upon the assumption that this domain is already affected in the prodromal period. Aiming to find RTT-specific speech-language atypicalities on signal level as early acoustic markers, we analysed more than 16 hours of home video recordings of 4 girls later diagnosed with RTT and 4 typically developing girls aged 6 to 12 months. We segmented a total of 4 678 pre-linguistic vocalisations. A comprehensive set of acoustic features was extracted from the vocalisations as basis for the classification paradigm RTT versus typical development. A promising mean unweighted recognition accuracy of 76.5% was achieved using linear kernel support vector machines and 4-fold leave-one-speaker-pair-out cross-validation. To the best of our knowledge, this is the first approach to automatically identify infants later diagnosed with RTT based on acoustic characteristics of pre-linguistic vocalisations. Our findings may build the basis for facilitating earlier identification and thus an avenue for an earlier entry into intervention.