Automatic intelligibility assessment and recognition of dysarthric speech using phonetic variations

Abstract

3) to provide effective speech recognition methods based on phonetic variation modeling

Research methods to be presented in this thesis are largely divided in the following three categories: 1) extraction of phonetic variation patterns from speech utterances of dysarthric individuals

2) development of speech intelligibility assessment algorithm based on phonologically-structured sparse linear model

3) development of speech recognition algo-rithm based on regularized speaker adaptation techniques in the framework of phonetic variation modeling. The above-mentioned our research efforts and findings will be valuable for researcher to make effective use of pho-netic variation information for the purpose of the intelligibility assessment and recognition of dysarthric speech.

Dysarthria is a motor speech disorder resulting from neurological injury of the motor speech system, im-peding the physical production of speech. Therefore, patients with dysarthria often have trouble in pronouncing certain sounds, resulting in undesirable phonetic variation

their speech intelligibility is reduced in proportion to the severity of dysarthria, and patients with dysarthria have difficulty in communicating with others. In addi-tion, dysarthria is often accompanied with a physical disability such as cerebral palsy that limits the speaker’s capability to communicate through computers and electronic devices. For those who suffer from dysarthria, automatic speech recognition (ASR) can help in controlling computers and electronic devices. However, current ASR systems for the general public are not well-suited to dysarthric speech due to the phonetic variation. Hence, it is necessary to develop an ASR system specialized for dysarthric speech. In general, the characteristics of dys-arthric speech is characterized by the speech intelligibility, and therefore, the automatic intelligibility assessment method of dysarthric speech can be useful in the fields of dysarthric speech processing. For example, it is possi-ble to apply appropriate speech recognition techniques depending on the intelligibility and automatic intelligibil-ity assessment can help speech therapists in diagnosing the degree of speech disorder. Thus, we focus on both automatic speech recognition and automatic speech intelligibility assessment that have been emerging to help people who suffer from dysarthria in this thesis. The main goals of this thesis are summarized as follows: 1) to demonstrate that phonetic variation in-formation can play an important role in the automatic intelligibility assessment and recognition of dysarthric speech

2) to propose effective intelligibility assessment algorithm based on the selection of phonetic variations