Multimode variable bit rate speech coding: an efficient paradigm for high-quality low-rate representation of speech signal

摘要

The speech signal consists of a time-varying ensemble of different types of segments with distinct characteristics, which require different degrees of coding resolution in order to retain an overall high voice quality. A fixed-rate coder can capture such time-varying characteristics only if it operates at a high enough bit rate. At a low bit rate, a fixed-rate coder will not be able to capture all of these various segments well and will fail to render high voice quality. A multimode variable bit rate (VBR) coder uses an arsenal of modes, operating at different bit rates. These modes are designed to represent these different speech segments optimally with the right amount of coding resolution. Thus, a multimode VBR codec adapts the coding mechanism to the input speech and delivers high quality at low (average) rates. This paper presents the essential framework and the unique advantages of a multimode VBR codec and suggests algorithms for the different modes.