Dysarthric speech synthesis via non-parallel voice conversion

In this thesis we propose and evaluate a voice conversion (VC) method to synthesise dysarthric speech. This is achieved by a novel method for dysarthric speech synthesis using VC in a non-parallel manner, thus allowing VC in incomplete and difficult data collection situations. We focus on two applications: First, we aim to improve automatic speech recognition (ASR) of people with dysarthria by using synthesised dysarthric speech as means of data augmentation. Unimpaired speech is converted to dysarthric speech and used as training data for an ASR system. The results tested on unseen dysarthric words show that the recognition of severe dysarthric speakers can be improved, yet for mild speakers, an ASR trained with unimpaired speech performs better. Secondly, we want to synthesise pathological speech to help inform patients of their pathological speech before committing to an oral cancer surgery. Knowing the sound of the voice post-surgery could reduce the patients' stress and help clinicians make informed decisions about the surgery. A novel approach about pathological speech synthesis is proposed: we customise an existing dysarthric (already pathological) speech sample to a new speaker?s voice characteristics and perform a subjective analysis of the generated samples. The achieved results show that pathological speech seems to negatively affect the perceived naturalness of the speech. Conversion of speaker characteristics among low and high intelligibility speakers is successful, but for mid the results are inconclusive. Whether the differences in the results for the different intelligibility levels are due to the intelligibility levels or due to the speakers needs to be further investigated