Author ORCID Identifier
Date Available
9-25-2022
Year of Publication
2022
Document Type
Doctoral Dissertation
Degree Name
Doctor of Engineering (DEng)
College
Engineering
Department/School/Program
Electrical and Computer Engineering
Advisor
Michael T. Johnson
Abstract
Dysarthria is a motor speech disorder often characterized by reduced speech intelligibility through slow, uncoordinated control of speech production muscles. Automatic Speech recognition (ASR) systems may help dysarthric talkers communicate more effectively. However, robust dysarthria-specific ASR requires a significant amount of training speech is required, which is not readily available for dysarthric talkers.
In this dissertation, we investigate dysarthric speech augmentation and synthesis methods. To better understand differences in prosodic and acoustic characteristics of dysarthric spontaneous speech at varying severity levels, a comparative study between typical and dysarthric speech was conducted. These characteristics are important components for dysarthric speech modeling, synthesis, and augmentation. For augmentation, prosodic transformation and time-feature masking have been proposed. For dysarthric speech synthesis, this dissertation has introduced a modified neural multi-talker TTS by adding a dysarthria severity level coefficient and a pause insertion model to synthesize dysarthric speech for varying severity levels. In addition, we have extended this work by using a label propagation technique to create more meaningful control variables such as a continuous Respiration, Laryngeal and Tongue (RLT) parameter, even for datasets that only provide discrete dysarthria severity level information. This approach increases the controllability of the system, so we are able to generate more dysarthric speech with a broader range.
To evaluate their effectiveness for synthesis of training data, dysarthria-specific speech recognition was used. Results show that a DNN-HMM model trained on additional synthetic dysarthric speech achieves WER improvement of 12.2% compared to the baseline, and that the addition of the severity level and pause insertion controls decrease WER by 6.5%, showing the effectiveness of adding these parameters. Overall results on the TORGO database demonstrate that using dysarthric synthetic speech to increase the amount of dysarthric-patterned speech for training has a significant impact on the dysarthric ASR systems.
Digital Object Identifier (DOI)
https://doi.org/10.13023/etd.2022.360
Recommended Citation
Soleymanpour, Mohammad, "SYNTHESIZING DYSARTHRIC SPEECH USING MULTI-SPEAKER TTS FOR DSYARTHRIC SPEECH RECOGNITION" (2022). Theses and Dissertations--Electrical and Computer Engineering. 184.
https://uknowledge.uky.edu/ece_etds/184
