Reduction of CPU Computation Load Based on OpenCL for Speech Codec

The G.723.1 codec has been widely used and implemented for real-time Voice over Internet Protocol (VoIP). The implementation should consider the computational power for real-time encoder. This paper aims to reduce the computational load of CPU-based G.723.1 codec implementation. The G.723.1 encoding can be parallelized using Graphical Processing Units (GUP) to further reduce the computation time. With the assistance of GPU handling, the CPU could have extra computing power to perform other operations. Nowadays, each computer or handheld mobile device will have a general low-cost graphics card with GPU. So using the GPU as an auxiliary device does not increase the hardware cost, but makes good use of the computer hardware resources. The overall evaluation results present that the average perceptual evaluation of speech quality score is slightly reduced by 0.012 and the method we proposed can reduce the computational loading of CPU by about 22.3% relative to the original CPU only. Furthermore, an objective speech quality assessment validates that the proposed method can provide comparable speech quality to the original coding method.