We propose an efficient codebook design method for algebraic CELP (ACELP) speech coder and an improved differential quantization method with algebraic codebook gain.
As the number of non-zero pulses of an algebraic codebook increases, the speech quality is improved but the computational load in searching the codebook is increased exponentially. The proposed codebook design method reduces the computational load in searching the codebook by implementing two algebraic codebooks which have different pulse positions. Moreover, since two algebraic codebook gains are highly correlated, the number of bits for transmitting codebook gains is reduced by transmitting only one codebook gain.
A differential quantization method is used to reduce the number of bits for transmitting algebraic codebook gain by adopting the algebraic codebook gain of current subframe which is slightly different from that of previous one. However, the error propagation problem is occurred when a differential quantization is used. In order to avoid this problem, a differential quantization method is not used to encode first subframe; it is used to encode second, third, and fourth subframe.
By adopting the proposed codebook design method and differential quantization method to ACELP speech coder, we improve the performance of ACELP speech coder. Two bits are saved during the transmission of a subframe by adopting a differential quantization to ACELP speech coder. The performance of the improved ACELP speech coder is investigated by computer simulation for various speech data and compared with MP-MLQ speech coder (ITU-T recommendation G.723.1). The simulation results show that the improved ACELP speech coder reduces codebook search time by a third of that of MP-MLQ speech coder and the subjective and objective speech quality of the improved ACELP speech coder and MP-MLQ speech coder are indistinguishable.