LAUSR

Feature extraction in speech signals under the influence of background excitation is a challenging task. In this research, we propose phoneme subspace integrated with the linear visual assessment tendency (LVAT) algorithm to retrieve the audio feature based on spectral depth analysis. LVAT algorithm performs a clustering of different spectral features to define the intensity of signal weight. The Fast Fourier transform (FFT) projects selection of weight estimated samples from the signal for phoneme subspace. The FFT-phoneme subspace combination enhances the feature by analyzing the low, middle and high-frequency signals based on phone subspace weight update. Traditional feature extraction techniques like mel frequency cepstral coefficients, linear predictor cepstral coefficients and power normalized cepstral coefficients are analyzed under different noise conditions and compared with the results of clustering with power normalized cepstral coefficients. The experimental results demonstrate improvement in the performance by comparing the objective measures such as sensitivity, specificity, accuracy and recognition rate.