Formant weighted cepstral feature for LSP-based speech recognition

In this paper, we propose a formant weighted cepstral feature for an LSP-based speech recognition system. The proposed weighting scheme is based on the well-known property of LSPs that the speech spectrum has a peak when adjacent LSFs come close. By applying this scheme to the pseudo-cepstrum (PCEP) conversion process (Kim et al. 1993), we can obtain formant weighted or peak enhanced cepstral features. Results of speech recognition experiments using QCELP coder output show that the proposed feature set outperforms the conventional features such as LSP or PCEP. Moreover its performance also exceeds that of the unquantized LPC cepstrum