This paper proposes a new objective quality assessment method for bit-reduction coding of wideband speech taking into account the masking effect of quantizing noise. First, this paper analyzes the reliability and sensitivity of the speech quality assessment method, based on a paired-comparison test with a modulated noise reference signal, for the bit-reduction coding of high-quality wideband speech. Then, the perception of quantizing noise is studied using speech with noise synthesized similar to the quantizing noise. The detection of quantizing noise is found to be influenced by masking by the source signal. This leads to a new method of objectively estimating the quality of coding speech by multiple regression analysis. The factors for the estimation are segmental signal-to-noise ratio, spectrum envelope distance between source signal and quantizing noise, and the similarity of the noise power envelope to the source signal in the time domain. This estimation method is applied to the parameter optimization of wideband coding systems.
[1]
B. Atal,et al.
Optimizing digital speech coders by exploiting masking properties of the human ear
,
1978
.
[2]
Nobuhiko Kitawaki,et al.
Objective quality evaluation for low-bit-rate speech coding systems
,
1988,
IEEE J. Sel. Areas Commun..
[3]
Ronald E. Crochiere,et al.
A study of complexity and quality of speech waveform coders
,
1978,
ICASSP.
[4]
Masaaki Honda,et al.
Bit allocation in time and frequency domains for predictive coding of speech
,
1984
.
[5]
John Makhoul,et al.
Adaptive noise spectral shaping and entropy coding in predictive coding of speech
,
1979
.
[6]
E H Rothauser,et al.
Isopreference method for speech evaluation.
,
1968,
The Journal of the Acoustical Society of America.