Comparison of optimal quantizations of speech reflection coefficients

Four quantization schemes for the reflection coefficients obtained from linear prediction speech analysis are theoretically compared. The asymptotic performance of each scheme is developed and compared using various definitions of "optimal." If the coefficients are treated as independent parameters, it is shown that a fixed bit rate minimum deviation quantization without noiseless source coding can realize within 0.26 bits per coefficient, the ultimate bit rate achievable by allowing noiseless source coding of the data.