Speech classification based on compressive sensing measurement sequence

Abstract

This paper designe a speech classifier based on compressive sensing (CS) measurement sequence which mainly solves the problem controlling robot and other intelligent equipments in the environment where we only need to use simple voice commands. In this paper, we obtain speech signal measurement sequence by using the row ladder matrix, then extract Mel Frequency Cepstral Coefficient (MFCC) matrix and reshape it into one-dimensional features. Moreover, we respectively obtain the classification model by using Supported Vector Machine (SVM) and K-Nearest Neighbor (KNN) algorithm. Experiments show that the highest accuracy rate of classifier up to 95% by using the KNN with Cosine distance model after that we tested for different distance models, and the accuracy rate is 91.95% by using the C-Support Vector Classification (C-SVC) of SVM classifier.