Abstract—This paper implements cepstral feature classification of compressive sensed speech signal and compares it with results obtained without compressive sensing. The Orthogonal Matching Pursuit algorithm is used for reconstruction with Gaussian as the sensing matrix and Discrete Cosine Transform as the sparsifying basis. The paper also uses post processing of compressive sensed signal to improve the accuracy of classification. The K-Nearest Neighbor classifier was tested for different distance models, with City Block giving the highest accuracy of 57.14% and 91.43%, with and without compressive sensing respectively. Post-Processing of compressive sensed signal using a median filter improves the accuracy from 57.14% to 80%.
 

Index Terms—compressive sensing, speech recognition, orthogonal matching pursuit, K-nearest neighbor, cepstrum

Cite: Usham V. Dias, Jeswil E. Mascarenhas, and Lioshka J. Dias, "Compressive Sensed Speech Recognition," International Journal of Signal Processing Systems, Vol. 4, No. 6, pp. 483-486, December 2016. doi: 10.18178/ijsps.4.6.483-486