An 8.0-/8.4-kbps wideband speech coder based on mixed excitation linear prediction

Abstract

A new method based on the Mixed Excitation Linear Prediction (MELP) model for coding wideband speech signals at medium bit rates is proposed. Modifications are made to the voiced/unvoiced strength analysis, pitch estimation and linear predictive (LP) analysis/synthesis of the Federal MELP Standard. An 8.0-/8.4-kbps wideband MELP coder is developed based on the modifications. Both the objective and subjective tests show that the subjective quality of the synthesized clean speech coded by the proposed 8.4-kbps wideband MELP coder is comparable to that of the ITU G.722 (P. Mermelstein, G. 722: a new CCITT coding standard for digital transmission of wideband audio signals, IEEE Comm. Mag. 26 (1) (1988) 8–15.) coder coding at 48 kbps. For the operating bit rate of 8 kbps, the performance of the coder is slightly better than that of MPEG4 CELP (Code-Excited Linear Prediction) (J.D. Johnson, S.R. Quackenbush, J. Herre, B. Grill, Review of MPEG-4 general audio coding, in: Multimedia Systems, Standards and Networks, A. Puri, T. Chen (Eds.), Marcel Dekker, New York, 2000 (Chapter 5)) coder operating at 14.4 kbps.