PCA GAUSSIAN MIXTURE MODEL FOR TWO-DIMENSIONAL ROOT CEPSTRUM FEATURES OF SPEECH

Feature extraction plays a substantial role in automatic speech recognition systems. The Two-dimensional root cepstrum (TDRC) is a special case of feature extraction for speech recognition which has some advantage over the others methods. The exact estimation of the model of density function of TDRC is an important stage in building a successful speech recognition system. A PCA Gaussian mixture model has been suggested for TDRC features of speech. Experimental results which have been done on TIMIT database show that the proposed model is a proper model for TDRC features