| 摘要 |
PURPOSE:To apply recovery and optimization by expressing an input signal with a signal where the mean square error due to a linear forecast parameter and the distortion of the signal itself corresponds directly the means square error between the input signal and a recovery signal thereby minimizing the quantized distortion of the signal itself for a quantizer. CONSTITUTION:A linear forecast analyzer 110 calculates a reflection coefficient ki being one of linear forecast parameters from N sets of voice samples, quantizes them and outputs the result to a linear forecast parameter coder 140 and a linear conversion computer 130. The computer 130 converts the reflec tion coefficient ki into a linear forecast coefficient ai to obtain a covariance matrix RH, then the covariance matrix RH and a diagonal matrix are decom posed. The obtained unitary matrix F and the diagonal matrix are outputted to a conversion coefficient computer 120. A signal series ff is generated from a voice signal series x(n), the matrix F and the diagonal matrix. After the signal series ff is coded by a coefficient coder 150, the result is multiplexed with a code representing a reflection coefficient by a multiplexer 160 and sent to a circuit network from a transmission output terminal 170. |
