Analysis of subband quantization noise level and shapes: a function of wideband audio codec tandemming

Newly developed codecs for wideband audio signals rely on subband coding and psychoacoustic modeling to achieve significant reduction in the bit rate of a digital audio signal. An investigation into the shape and level of quantization noise introduced into a coded audio signal is performed. Psychoacoustic principles reveal that quantization noise is masked, i.e. made inaudible, if its spectral shape fails below that of the original signal by a 13 dB threshold. Simulation results show that there is a strong correlation of quantization noise and original signal shapes at low bit allocations. Quantization noise levels which increase result in a degradation of the quality of a codec. A criterion for the evaluation of a codec is its performance when tandemmed. Simulation results reveal that there can be up to an 18 dB increase in average subband noise energy for eight tandems. Surprisingly, subbands more vulnerable to an increase in quantization noise tend to exhibit a high signal to mask ratio, as determined by the psychoacoustic model