Sparse Coding has been an active research topic in machine learning and signal processing for the last ten years, as it has achieved impressive results when applied to many problems such as face recognition and image denoising. In this paper, we present a new contribution in applying sparse coding to the problem of Arabic phoneme classification. The classification system which is entitled: Sparse Coding based phoneme Classification system (SCPCS), employs the sparse code as a new speech feature for classification using Sparse Representation Classifier. The Sparse code is simply the “coefficients” of the “sparse” (with many zeros) linear combination of basic signals that can represent the targeted signal as close as possible. We study the impact of the sparse coding solver which aims to produce the sparse code, on its discrimination capability. Experiments to evaluate the proposed system performance were conducted on two manually segmented Arabic phonemes, extracted from KAPD (King Abdulaziz city for science and technology Arabic Phonetic Database) and CSLU2002 (Centre for Spoken Language Understanding) Arabic speech databases. Experimental results showed that the proposed system has achieved an accuracy of 85.3% on KAPD and 53.4% on CSLU2002, which are better than state of the art results in these two datasets.
International Journal of Engineering Trends and Technology.
2017.
Students
Teachers
Graduates
Faculties
