An Experimental 9600-bits/s Voice Digitizer Employing Adaptive Prediction

An experimental model of a coder for transmission of speech over a 9600-bits/s digital channel was built to demonstrate feasibility of an adaptive prediction-coding technique. After analog-to-digital conversion of the speech input, the coder employs digital processing using a computer type organization. Resonances in the short-term speech spectrum are removed by a nonrecursive digital transmit filter and the resulting uncorrelated signal is coded by an 8000-bits/s direct feedback delta coder. The transmit filter parameters are adapted to the input spectrum by a least squares algorithm involving calculation of short term correlation coefficients of the sequence of input samples. These filter parameters are multiplexed with the delta coder output for transmission to the receiver. A recursive receive filter restores the original speech spectrum. A computer simulation of the voice digitizer was performed to determine the order of the digital filters and to optimize other parameters prior to the design of the experimental model. The results of the simulation and design considerations for the experimental model are described.