Speech coding technologies for compressing digital speech signals are being explored to help better utilize limited bandwidth resources and to support communications privacy. This article reports on recent developments at Mitsubishi Electric onlow bit-rate speech coding technology which is central to cellular telephones and has numerous other applications.Speech coding technology provides the voice bandwidth compression used in cellular telephone systems worldwide. The compressed signal has a bit rate of 3.45~ 13kbps, which increases to 5.6~ 22.8kbps when error correction is added. Most of these methodsemploy some type of the code excited linear prediction (CELP) algorithm illustrated in Fig. 1. This algorithm involves dividing the speech signal into 20~ 40ms frames, and then representing each frame by the best approximation of sound vectors from anexcitation codebook. Only the indices of these vectors are actually transmitted. Although codebooks can be designed to capture the characteristics of speech, extensive calculations are needed to search for the best mix of vectors to express speech at lowbit rates. Studies on codebooks and search methodologies are being conducted to reduce the computational requirements and increase the sound quality, while other studies are being conducted on increasing system robustness toward channel errors andbackground noise.
展开▼
