Enhanced polynomial kernel (EPK)–based support vector machine (SVM) (EPK-SVM) classification technique for speech recognition in hearing-impaired listeners

摘要

Automatic speech recognition of Tamil Language with Hearing-Impaired becomes difficult task in recent decades. In order to deal with the challenges with speech perception in hostile listening situations, Noise Reduction (NR) algorithms have been developed with the aim of improving the speech intelligibility (SI), speech quality, and ease of listening. Even though the noises are removed, extraction of correct features from the speech becomes difficult task. The major aim of this work is to introduce a Classification Technique for Speech Recognition in Hearing-Impaired Listeners. The binary mask along with its binary weights and the Wiener filter with constant weights form the representatives of a hard and a soft-decision scheme for time-frequency masking. In the proposed Log Frequency Power Coefficients (LFPC), Pitch, Mel-Frequency Cepstral Coefficients (MFCCs), Energy, formants, and intensity as input feature vectors are extracted from preprocessed signals. Then, for automatic speech recognition process, Enhanced Polynomial Kernel (EPK)-based Support Vector Machine (SVM) (EPK-SVM) classifier is proposed for Hearing Impaired in Tamil language is implemented in MATLAB software. The results obtained showed that SVM is found to be potential in hearing-impaired application and is validated via the use of the recognition accuracy and error rate.