High quality mid-rate speech coding

The authors present the results of a feasibility study comparing two basic classes of coders in the mid-rate range (6-8 kbps): CELP and sinusoid based coders. It is concluded that CELP coding is a mature technique able to yield high-quality synthetic speech with some background noise at the mid-rate range. It has a tractable computational complexity and good robustness, making it the best present candidate. Postprocessing can enhance its output quality, though further research is needed in this field. Sinusoidal coding, together with the use of narrowband basis functions, can produce synthetic speech of higher quality than that produced by CELP. Its robustness however, must be increased. This work suggests that a significant research effort into this technique is fully justified.<<ETX>>