Evaluation and Optimization of Nonlinear Quantization Characteristics in PCM Transmission of Speech
Hiroya Fujisaki, Masanori Kataoka · The Journal of the Acoustical Society of America · 1967
Nonlinear quantization has been introduced into PCM transmission of speech in order to reduce the quantization effects. Although signal-to-noise ratio has been conventionally adopted for the evaluation of such systems, assuming a negative exponential distribution for speech amplitude, there is little evidence for its validity. An evaluation theory is presented based on the information transmission rate of a quantized channel and approximating the temporal structure of speech by two distributions corresponding to voiced and unvoiced sounds. Articulation tests have been performed using both actual PCM systems and a flexible simulator and the results have been analyzed both in terms of the articulation score and the amount of information transmitted per syllable. The results indicate poor correlation between conventional calculations and articulation scores, and prove the validity of the proposed theory. Comparisons of known nonlinear characteristics lead to an optimum one that is somewhere between the logarithmic and the hyperbolic characteristics.