New approach to speech digitization combining time-domain harmonic scaling and adaptive residual coding

This report describes the results of a twelve-month study, supported under DCA Contract 100-80-C-0050, of a new speech digitization algorithm combining Time-Domain Harmonic Scaling (TDHS) and Adaptive Reidual Coding (ARC). The FORTRAN simulation of this system conducted as part of this study produces high quality speech reproduction at mediumband bit rates of 9.6 kb/s and 16 kb/s. This system also displays excellent robustness characteristics for channel bit error rates as high as 1% and for acoustic background noise. By basing the required pitch extraction on a three-level clipped signal, the hardware requirements for the system are kept modest. The combined algorithm has several features which become significant in a fuel system applications. Because the algorithm is a high performance, waveform matching algorithm, extremely good performance in the random configuration with other algorithms is anticipated. Since the technique is basically a waveform reconstruction technique, it will perform well on non speech signals such as in-band signaling and modem tones.