Exploiting ultrasound tongue imaging for the automatic detection of speech articulation errors

Speech sound disorders are a common communication impairment in childhood. Because speech can negatively affect the lives and development of children, clinical intervention is often recommended. To help with diagnosis treatment, clinicians use instrumented methods such as spectrograms or ultrasound tongue imaging to analyse articulations. Analysis these be laborious for clinicians, therefore there growing interest its automation. In this paper, we investigate contribution automatic detection articulation errors. Our systems trained on typically developing child augmented database adult using audio ultrasound. Evaluation indicates that pre-training jointly gives best results an accuracy 86.9%. evaluate disordered speech, collect pronunciation scores from experienced language therapists, focusing cases velar fronting gliding /r/. The show good inter-annotator agreement fronting, but not For error detection, obtained when audio. system correctly detects 86.6% errors identified by clinicians. Out all segments system, 73.2% match Results harder interpret due poor agreement, appear promising. Overall findings suggest has potential integrated into software automatically quantifying progress during therapy.