VARIABLE-RATE CELP BASED ON SUBBAND FLATNESS

Code-excited linear prediction (CELP) is the predominant methodology f or communications quality speech coding below 8 kbps, and several vari able-rate CELP schemes have been discussed in the literature, includin g QCELP, the variable-rate wideband digital cellular mobile radio spee ch coding standard specified in IS-95. A key component of these speech coders is the detection and classification of speech activity, and se veral cues for rate variation have been studied, such as measuring sho rt-term speech energy, deciding whether the speech is voiced or unvoic ed, or making more sophisticated phonetic classifications. We present a new method for rate variation based on a measure of subband spectral flatness, called spectral entropy. Spectral entropy is a normalized i ndicator of the texture of the input spectrum and is thus less depende nt on speech and background noise energy variations. We present some r esults on the use of spectral entropy for voice activity detection acr oss subbands and then evaluate using spectral entropy for deriving mod e and rate allocation cues for a variable-rate CELP coder operating at an average rate of 2 kbps. To achieve communications quality speech a t this rate, we develop a new split-band vector quantization (VQ) tech nique for representing the line spectral pairs and a multiple codebook approach for efficiently quantizing the coefficients of a three-tap p itch predictor, called lag-indexed VQ.