| 摘要 |
PURPOSE:To perform faster voice coding by reducing computing operation during evaluating errors while minimizing the degradation of synthesized voice. CONSTITUTION:A driven vector 102 is selected by a code book memory means 101. A voice synthesizing means 103 synthesizes voice in accordance with linear prediction analysis by using the driven vector 102 as an input element. A thinning-out mean 106 performs individual thinning-out of the element value of an input voice vector 105 and the element value of a synthesized voice vector 104 at certain intervals. An error evaluating means 109 performs square error evaluation of a thinned-out input voice vector 107 and a thinned-out synthesized voice vector 108. The above processings are repeated for every driven vector 102 so that the input voice vector 105 is coded in accordance with parameters which brings the synthesized voice vector with the minimum square error (maximum correlation value). |
