VA Research and Development LOGO

Logo for the Journal of Rehab R&D
Volume 42 Number 3, May/June 2005
Pages 363 — 372


Abstract - Effect of high-frequency spectral components in computer recognition of dysarthric speech based on a Mel-cepstral stochastic model

Prasad D. Polur, PhD;* Gerald E. Miller, PhD

Department of Biomedical Engineering, Virginia Commonwealth University, Richmond, VA
Abstract — Computer speech recognition of individuals with dysarthria, such as cerebral palsy patients, requires a robust technique that can handle conditions of very high variability and limited training data. In this study, a hidden Markov model (HMM) was constructed and conditions investigated that would provide improved performance for a dysarthric speech (isolated word) recognition system intended to act as an assistive/control tool. In particular, we investigated the effect of high-frequency spectral components on the recognition rate of the system to determine if they contributed useful additional information to the system. A small-size vocabulary spoken by three cerebral palsy subjects was chosen. Mel-frequency cepstral coefficients extracted with the use of 15 ms frames served as training input to an ergodic HMM setup. Subsequent results demonstrated that no significant useful information was available to the system for enhancing its ability to discriminate dysarthric speech above 5.5 kHz in the current set of dysarthric data. The level of variability in input dysarthric speech patterns limits the reliability of the system. However, its application as a rehabilitation/control tool to assist dysarthric motor-impaired individuals such as cerebral palsy subjects holds sufficient promise.
Key words: cerebral palsy, communication disorder, dysarthric speech, hidden Markov model, high-frequency spectral components, human-machine interfaces, Mel-frequency cepstral coefficients, rehabilitation, speech and motor disabilities, speech recognition.

 → go to Contents Page for Volume 42, No 1
 → go to HTML version of this article
 → go to PDF version of this article