Time-varying discriminative feature extraction effective for phonetic discrimination

We investigate the interactions of front-end preprocessing and back-end classification techniques in hidden Markov model (HMM) based speech recognition. The proposed model aims at finding an optimal linear transformation on the mel-warped discrete Fourier transform (DFT) with the construction of dynamic feature parameters according to the minimum classification error (MCE) criterion. Experimental results show that state-dependent transformation on mel-warped DFT features is superior in performance to the mel-frequency cepstral coefficients (MFCCs). An error rate reduction of 15% is obtained on a standard 39-class TIMIT phone classification task in comparison with the conventional MCE-trained HMM using MFCCs and delta MFCCs that have not been subject to optimization during training.