Speech recognition and classification using the compressive sensing method

The research is aimed to present speech signal recognition and classification using the compressive sensing method to reduce the data from speech signal cross-correlation and to compare similarity and difference between the reference speech signal (the researcher's signal) and comparative signals. The speech signal reconstruction method is based on the solution to the underdetermined linear inverse problem. The number of equations measured is less than that of data dimensions (m<;<;n). This is solved by the Tikhonov regularization. The research efficiency of speech recognition and classification is presented by comparing norm values and regularization values (λ) normwise relative errors and regularization values (λ) the ratio of ℓ
₁
/ ℓ
₂
norms and regularization values (λ) to acquire the zero norm (ℓ
₀
=||x||
₀
) that means numerical sparse measurement. The research also indicates the effectiveness of speech signal reconstruction based on measurement of signal to noise ratio (SNR) between the reference speech signal (the researcher's speech signal), comparative speech signals and regularization values (λ). The benefit of the research will be applied to classifying individual speech signals for security.