Acoustic Analysis for Human Voice Disorder Classification Using Optimization and Machine Learning Techniques

Abstract

The diagnosis of voice disorders through aggressive medical techniques seems painful for patients. Hence, automatic speech recognition and disorder identification methods have drawn much interest in the recent years and have proved to be successful. In this work, voice recordings are taken from the Saarbruecken Voice Database. The signals are preprocessed to remove silence and de-noised using Hybrid Wiener Filter Discrete Wavelet Transforms (HWFDWT). Features are extracted using Cat Swarm Optimization Mel Frequency Cepstrum Coefficients (CSOMFCC). Finally, the features are classified using Classification using Modified Optimized Back Propagation Network Disorder voice Classification (MOBPNDC). The classification scheme outperforms the existing Support Vector Machine (SVM) and Back Propagation Neural Network (BPNN) methods in terms of Accuracy, Precision, Recall, F-Measure and Time period. newlinei) Major objectives : newlineTo develop an automatic voice pathological identification system based on vocal parameters to make the decision about the speech sample as Pathological or Normal using various speech features and machine learning classifiers. newlineiv) Findings: newlineand#61623; Proposed the Hybrid Wiener Filter Discrete Wavelet Transforms (HWFDWT) for Silence Removal and Noise Removal in Pre-processing techniques. newlineand#61623; Proposed the Wavelet Thresholding Algorithm for Silence Removal and Noise Removal. newlineand#61623; Proposed Cat Swarm Optimization Mel Frequency Cepstrum Coefficients (CSOMFCC) to extract the best features from the signal in the Feature Extraction process. newlineand#61623; Proposed the Modified Optimized Back Propagation Network Disorder Voice Classification (MOBPNDVC) for Voice Pathology Identification System construction. newlineand#61623; Proposed MOBPNDVC Optimization Algorithm for Disorder Voice classification accuracy calculation. newlineand#61623; Analyzed the Voice Pathological Identification System to identify five types of voice pathology such as Laryngitis, Laryngoceles, Dysphonia, Diplophonia, and Chorditis