Bandwidth Compression of Voice and Modem Signals using Variable-Rate ATC Encoding

We present a strategy based on local SNR for variable-rate encoding of a variety of telephony signals using Adaptive Transform Coding. We first describe simple modifications of the speech version of ATC, involving chiefly the side information quantization and the bit allocation procedures, which substantially improve fixed-rate encoding performance on modem signals. Then, using simple statistical models of bit-rate demand, we predict the compression performance of the variable-rate encoder in a typical telephony network. We compare the performance of variable-rate and fixed-rate encoding of speech and modem signals, based on extensive listening tests and simulations, respectively. We also report on the progress of an experimental implementation of the algorithm in 32-bit floating-point hardware.