Vocal fold disorder detection based on continuous speech by using MFCC and GMM

TLDR: Mel-frequency cepstral coefficients (MFCC) are used with Gaussian mixture model (GMM) to build an automatic detection system capable of differentiating normal and pathological voices.

Abstract: Vocal fold voice disorder detection with a sustained vowel is well investigated by research community during recent years. The detection of voice disorder with a sustained vowel is a comparatively easier task than detection with continuous speech. The speech signal remains stationary in case of sustained vowel but it varies over time in continuous time. This is the reason; voice detection by using continuous speech is challenging and demands more investigation. Moreover, detection with continuous speech is more realistic because people use it in their daily conversation but sustained vowel is not used in everyday talks. An accurate voice assessment can provide unique and complementary information for the diagnosis, and can be used in the treatment plan. In this paper, vocal fold disorders, cyst, polyp, nodules, paralysis, and sulcus, are detected using continuous speech. Mel-frequency cepstral coefficients (MFCC) are used with Gaussian mixture model (GMM) to build an automatic detection system capable of differentiating normal and pathological voices. The detection rate of the developed detection system with continuous speech is 91.66%.