Improving Dysarthric Speech Intelligibility Using Cycle-consistent Adversarial Training
Abstract
Dysarthria is a motor speech impairment affecting millions of people. Dysarthric speech can be far less intelligible than those of non-dysarthric speakers, causing significant communication difficulties. The goal of our work is to develop a model for dysarthric to healthy speech conversion using Cycle-consistent GAN. Using 18,700 dysarthric and 8,610 healthy control Korean utterances that were recorded for the purpose of automatic recognition of voice keyboard in a previous study, the generator is trained to transform dysarthric to healthy speech in the spectral domain, which is then converted back to speech. Objective evaluation using automatic speech recognition of the generated utterance on a held-out test set shows that the recognition performance is improved compared with the original dysarthic speech after performing adversarial training, as the absolute WER has been lowered by 33.4%. It demonstrates that the proposed GAN-based conversion method is useful for improving dysarthric speech intelligibility.
Cite
@article{arxiv.2001.04260,
title = {Improving Dysarthric Speech Intelligibility Using Cycle-consistent Adversarial Training},
author = {Seung Hee Yang and Minhwa Chung},
journal= {arXiv preprint arXiv:2001.04260},
year = {2020}
}
Comments
To be Published on the 24th February in BIOSIGNALS 2020. arXiv admin note: text overlap with arXiv:1904.09407