| The speech coding is the process of converting speech into digital bit streams for efficiently digital storage. With the bandwidth in communications becoming largely broadened, people will no longer be satisfied with just the clearness of the voice in video and audio services. More information about the speaker should be provided, like background music. That is to say, people are pursuing face-to-face speech quality. Consequently, the bandwidth(300Hz-3400Hz) of the conventional speech coding is not enough for providing high quality reconstructed speech, and thus the research on wideband(50Hz-7000Hz)speech coding is necessary.Speex is an open-source codec based on the Code-Excited Linear Prediction (CELP) algorithm. It is targeted mainly towards voice over IP (VoIP) applications so it is designed to be robust to lost packets. Its encoding is very flexible, according to different application environments will adopt a unified code stream format and coding algorithms to achieve multi-bit rate, multi-flexible sample-rate speech coding and adapt to the needs of voice communications networks. Speex supports multiple bit-rates, ranging from 250bps to 24.6 kbps in narrowband (8 kHz) operation and from 3.95kbps to 42.2 kbps in wideband (16 kHz) operation.This paper investigates on the Speex algorithm in depth, and proposes three kinds of algorithm-level optimization methods based on the shortcoming of large time complexity. Simulation results demonstrate that the time complexity of Speex algorithm can be reduced significantly when PESQ score has minimal effect. Then, above optimization algorithms are realized on the 24-bit fixed-point AR1688 chip. Besides, through code-level optimization, the whole program's computational complexity is decreased and the coding efficiency is improved dramatically. Final results show that the performance of Speex algorithm is good on AR1688 chip. |
PDF Full Text Request