Adaptive transform coding incorporating Time Domain Aliasing Cancellation

Abstract

Adaptive transform coding incorporating Time Domain Aliasing Cancellation (TDAC) is presented in this paper for speech coding at low data rates. The TDAC transform differs from the conventional Discrete Cosine Transform (DCT) in that the analysis/synthesis window length can be longer than the system decimation factor. The channel bandwidth of the TDAC transform is narrower when compared to the DCT and therefore results in less spectral interaction occurring between bands. The TDAC transform coder employs forward adaptive scalar quantization of the transform coefficients and dynamic bit allocation over a block time corresponding to three transform operations. The associated side information is vector quantized, resulting in a side information overhead of 1.625 K bits per second. Both subjective and objective comparisons were made between TDAC and DCT coders at data rates of 9.6 K, 12 K and 16 K bits per second. The objective comparison involved segmental signal to noise ratio (SNRSEG) measurements while the subjective quality assessment was determined from A-B comparisons of recovered speech passed through the TDAC coder, the DCT coder and log PCM coders. The results indicate the TDAC transform coder is both objectively and subjectively superior to a transform coder employing the conventional DCT.