A Variable-Band Coding Scheme or Speech Encoding at 4.8 kb/s

Recently, a method for digitally coding speech signals in terms of sub-bands of the total spectrum was introduced that resulted in an improvement in quality of the coded signal over that obtained from a single full-band coding of the total spectrum. 1,2 The rationale for coding the signal in sub-bands is based upon the experimental fact that quantizing distortion is not equally detectable at all frequencies, and hence, the quality of the coded signal can be significantly improved by controlling the distribution of quantizing noise across the signal spectrum. Coding the signal in sub-bands offers the possibility of achieving this control. In the recent work by Crochiere, Webber, and Flanagan, 1 ' 2 the selection of the appropriate sub-bands was guided by the perceptual data contained in the so-called articulation index (Al). 3 The articulation index denotes, on the average, the contribution of each part of the spectrum 771 to the overall perception of the spoken sound. For high-quality speech at moderate bit rates (16 kb/s and greater), the frequency range 200 to 3200 Hz was partitioned into four fixed contiguous bands that contributed equally to the AI. The transmission bit rate of the sub-band coder could be lowered gracefully by limiting these sub-bands in width and by tolerating some spectral gaps that did not contribute significantly to the AI. However, carried to excess, the noncontiguous bands produced an unpleasant, reverberant quality in the signal that finally resulted in unacceptable speech.