Improvement in speech quality due to time incoherence of quantization noise in separate frequency bands
Paul G. Mermelstein · The Journal of the Acoustical Society of America · 1979
New speech coding techniques such as sub-band or splitband coding use independent quantizers to encode the speech information in separate frequency bands. This work is intended to improve our understanding of the perceptual effects of independent quantizers operating on separate frequency bands. The subjective preference of multiplicative white noise introduced independently into two 2-kHz wide speech bands has been compared to that of fullband 0–4 kHz noise of equal overall energy. In terms of both the just noticeable noise level, as well as the noise level perceptually equivalent to full-band noise at signal-to-noise ratios of 20 and 30 dB, introducing multiplicative noise independently into the two bands achieves an SNR gain of 2–5 dB. These results suggest that a significant advantage results from splitting the quantization noise into the separate bands because the time-coherence of the noise components is thereby to a large extent destroyed. This advantage is additional to that derived from the lower signal-to-noise ratio tolerance in the 2–4 kHz frequency band compared to the 0–2-kHz band. [Research supported by Grant A0105 from the National Research Council, Canada.]