Discriminating Parkinson and Healthy People Using Phonation and Cepstral Features of Speech

Savitha S. Upadhya, Alice N. Cheeran

Procedia Computer Science · 2018 · 28 citations · 10 references

DOIFull text

Open access

Concepts

Abstract

Parkinson’s disease is a neurological disease associated with the dopamine deficiency which affects the muscular movements of the body. The disease also affects the muscles used in the production of speech thereby altering the different features of speech. The analysis of these altered speech features will thereby aid in detecting the disease. In this paper the investigation is done to classify Parkinson diseased and healthy people by extracting the phonation and cepstral features of speech. The fourteen features of phonation analyzed includes five jitter variants, six shimmer variants, two features of harmonicity and the mean autocorrelation of the fundamental frequency feature. The cepstral features extracted are the twelve Mel frequency cepstral coefficients (MFCC). These features extracted from speech samples of both the groups is then classified using a neural network classifier and their performance is evaluated in terms of the different performance metrics namely recognition accuracy, equal error rate, sensitivity and specificity. The performance metrics is evaluated by conducting the experiment on only the phonation features followed by only the cepstral features and then using hybrid of both the features. The results obtained shows that the metrics of performance is found to be maximum using the phonation features. The recognition accuracy obtained using only the phonation features is 98.0 %, using only the cepstral features is 81.1 % and hybrid is 96.7 %.

References

10