Enhancement of dysarthric speech for developing an effective speech therapy tool

Dysarthria is a neuromuscular disorder that results from weakened movement of muscles used in speech production. This results in poor articulation causing the dysarthric speech to be slurred and difficult to understand. The natural auditory feedback makes the patients understand that their speech is of low quality, and this lowers their self confidence and they become more and more introverted causing the disorder to aggravate. In this work, we enhance the dysarthric speech and provide the enhanced speech to the patient through auditory feedback. This helps the patients to feel comfortable with their speech and gradually develop confidence to speak more and hence achieve a speedy rehabilitation. The utterances are analyzed using linear predictive coding (LPC). The LPC features in the acoustic space of dysarthric speaker are mapped to the feature space of the normative population using constrained maximum likelihood linear regression (CMLLR) before they are used for re-synthesising the enhanced speech. We then evaluated the quality of the enhanced speech using subjective and objective measures, DMOS and PESQ, and obtained an improvement of 63% and 43.4% for DMOS and PESQ measures respectively. Clinical trials are being pursued at Amrita Institute of Medical Sciences, Kochi on patients with dysarthria to evaluate the effectiveness of the proposed approach for faster rehabilitation of the patients. The results of these clinical trials will be reported in due course of time.