Low Complexity Fast Search Method for MP-MLQ and ACELP in G.723.1 Codec

Abstract

Due to the limited computational capability of personal information machines in multimedia communication, a coder with low computational complexity is necessary for integrating services from several media sources. The G.723.1 speech coder has been used extensively for Internet teleconferences. This paper presents two fast search methods to simplify computation of the excitation codebook search. For the multi-pulse maximum likelihood quantization (MPMLQ) codebook search, we propose an algorithm that uses some larger discrete cosine transformation (DCT) coefficients of the target signal to predict the candidate pulses. In addition, we redesign the MP-MLQ codebook to a twin multi-track position architecture. For the algebraic code excited linear predictive (ACELP) codebook search, we similarly use some larger DCT coefficients of the target signal to predict the candidate pulses. Simulation results indicate that the average Perceptual Evaluation of Speech Quality score is degraded slightly, by 0.017 and 0.021, relative to the original MP-MLQ and ACELP search procedures, respectively. However, we propose methods that can reduce the average computational complexity by approximately 45% relative to the original G.723.1 encoder computation load with perceptually negligible degradation. Objective and subjective evaluations verify that the two fast search methods we propose can provide speech quality comparable to that of using the original coder approach.