A multi-mode variable rate speech coder for CDMA cellular systems

This paper presents a multi-mode variable rate speech coder based on the CELP algorithm. The coder operates at a rate of 8.5 kbps, 4 kbps or 0.8 kbps with a 20 ms frame. The coder consists of five coding modes applied to distinct speech features. One out of the five coding modes is selected for each frame by using a mode selector which comprises a neural network and a speech power variation detector. To improve the coding performance, an inter-frame predictive LSP quantizer and a coding strategy for speech onsets are utilized. In low bit-rate speech coding, decoded speech quality is severely degraded in high background noise. A noise suppressor based on the spectral subtraction algorithm is also introduced in order to reduce background noises. We proposed this coder as a candidate for a new speech service standard of the North American CDMA digital cellular system IS-95. As a result of the first evaluation conducted by the Telecommunications Industry Association, the coder got the highest overall score on the tests in speech quality and average data rate.