Comparison of Feature Extraction Methods Between MFCC, BFCC, and GFCC with SVM Classifier for Parkinson’s Disease Diagnosis

Analysis of the voice signal can assist in the detection of Parkinson's disease, a degenerative and progressive neurological disorder affecting the central nervous system. Indeed, early changes in voice patterns and characteristics are frequently observed in patients with Parkinson's disease. Therefore, voice feature extraction can aid in the early identification of Parkinson's disease. This paper presents novel approaches to extracting features from speech signals using Gammatone frequency cepstral coefficients (GFCC), Bark frequency cepstral coefficients (BFCC), and Mel frequency cepstral coefficients (MFCC). The PC-GITA and Sakar databases are used, which contain speech signals from healthy individuals and individuals with Parkinson's disease. The coefficients from 1 to 20 of GFCC, BFCC, and MFCC are extracted from each speech signal and calculated the average value to extract the voiceprint of each speech signal. For classification, the support vector machine with different kernels (linear, RBF, and polynomial) and tenfold cross-validation are employed. Using the first 12 coefficients of the GFCC with a linear kernel yields a higher accuracy rate of 81.58% for Sakar database and 76% for PC-GITA database.