Scalable wideband speech coding using G.729 as a component

Abstract

SUMMARYThis paper proposes a scalable band-split wideband speech coding system that uses G.729 as a component and can decode speech from even part of the 16 kbit/s encoded data. The configuration of the proposed scalable encoding is as follows. The 7-kHz band speech signal is split into lower and upper bands, which are encoded separately. The lower band is scalably encoded at 12 kbit/s, and the upper band is encoded by CELP composed only of the noise codebook (4 kbit/s). In the lower band, the lower-level encoder (core layer) is included in the upper-level encoder (enhancement layer) as part of the functional element, to improve the performance of the whole lower-band encoder. To further improve quality while retaining scalability to the core coder, an additional pitch prediction scheme is proposed in which the excitation component of the enhancement layer is used in the pitch component. Performance evaluation by objective measures and a subjective evaluation experiment for the proposed schemes, showed that the quality is greatly improved compared with the conventional method.