B.S.T.J. Briefs: Adaptive Aperture Coding for Speech Waveforms - II

The quality of speech output in adaptive aperture coding1 has been improved by two refinements: (i) a code selection procedure based on formal error minimization rather than observations of aperture crossing,1 and (ii) a simple adaptive low-pass filtering operation based on adjacent sample correlation values, measured on a shortterm basis (typically, once every 20 to 30 ms). We describe these refinements with special reference to a 7-code aperture characteristic designed for an average output rate of 1.2 bits/sample, and speech inputs sampled at 8 and 12 kHz. At corresponding bit rates (9.6 to 14.4 kb/s), adaptive aperture coding, in conjunction with a first-order adaptive predictor, constitutes a medium-complexity approach in timedomain coding, with an output speech quality that is less-than-toll but nevertheless useful in many applications. A natural application of aperture coding is for speech storage where variability of output bit rate is less objectionable than in transmission. Adaptive aperture coding is a medium-complexity approach to the digitization of slowly changing waveforms. In a recently described1 procedure, the idea was to form an aperture centered on the last encoded waveform sample and to avoid further encoding until the waveform crossed that aperture. The features of the system that made it applicable to low bit rate digitization of speech were three. The first feature was an arrangement that precluded the need for explicit encoding of aperture crossing times.