Waveform speech coding based on vector transform quantization

Abstract

This paper presents a waveform speech coding system based on vector transform quantization (VTQ). VTQ is a coding system where consecutive M samples of a waveform are transformed into a set of coefficients which are quantized by a set of m < M vector quantizers (VQs). A coding gain higher than for the known scalar transform coding systems is achieved by employing vector quantization rather than the usual scalar quantization. Moreover, a bit allocation closer to the optimal one may be achieved by vector quantizers as a result of improved round‐off conditions. Assuming stationary input, the optimal bit allocation for the vector case is derived, and the coding gain achieved by such a system is evaluated. Techniques for adapting the bit assignment to the changing speech statistics are presented. These techniques are specific for the vector quantization approach and achieve adaptation along time and frequency axes. System simulation results are presented for typical speech waveforms at rates of 1–2 bits/sample.