This site will look much better in a browser that supports web standards, but it is accessible to any browser or Internet device.

Skip Navigation skip menu and banner
University of Wyoming

UW Technologies Available for Licensing

 

Technology Disclosure: Dysarthria


Dysarthria is a neurological disorder that hampers speech production.  This disorder affects both adults and children and is often a symptom of a disease, such as Parkinson’s disease or Cerebral Palsy, or the result of head trauma.  After a stroke almost a third of victims will have speech difficulty that can be characterized as dysarthria.  Dysarthria can be frustrating for the both the speaker and listener.

 

07-021: Design of Dysarthria Classifier Using the Global Statistics of Speech


There are five main types of dysarthria.  Each type is designated by the location of the injury in the person’s nervous system.  When there is a combination of two or more different types of dysarthria present it is classified as a sixth type, mixed dysarthria.  Dysarthria can be very frustrating for an inexperienced clinician to diagnose, but it is important to properly diagnose because the treatments vary among the different types. Researchers at the University of Wyoming have found a relationship between the type of dysarthria and its speech characteristic.  They have used this information to design a dysarthria classifier using global statistics (mean, variance, etc.) of features such as Mel-cepstrum coefficients (MFCC), perceptual linear predictions (PLP), and their delta and acceleration values.  

The classifier can distinguish between three types of dysarthria (Flaccid, Spastic, and Amyotrophic) that are commonly confused.  In order to maximize separation of the classes, features are selected from a large pool using a k-nearest neighbor probability density function (PDF) classifier.  These features are then used in a decision tree-like classifier that can identify the dysarthria type.  The accuracy with this classifier is 89.5%, it is non-intrusive, and no expertise is required to handle it.

 

07-022: Estimation of the Number of Syllables Using Hidden Markov Models

 
Speech rate measurement has applications in automatic speech recognition systems as well as in clinical settings, and automatic speech recognition can assist people suffering from dysarthria.  Researchers at the University of Wyoming have developed a computer-based method to estimate the number of syllables in a sentence, which can be used to estimate the speech rate of a person. 

The number of syllables is estimated by estimating the number of vowels in a sentence using hidden Markov models (HMMs).  HMMs are tools that model a signal statistically and are widely used in speech recognition applications.  The speech signal is divided into small segments and features such as Mel-cepstral coefficients are computed to form the observations.  The discrete HMM model parameters are estimated and the Viterbi algorithm is used to generate the most likely state sequence of the observations. 

 This method has been tested to count syllables in a sentence and also in diadochokinetic speech that contains repetitions of one syllable only.  An advantage of the HMM approach to estimate the number of syllables is that it is text-independent and can be used for speech segments of any duration.  Clinicians working with diadochokinetic rate will also benefit from this method as it requires little human intervention. 

If you would like to learn more about these novel inventions and how your company may apply them in commercial situations, please contact the director of the University of Wyoming Research Products Center, Davona Douglass.  We would be pleased to share further details.