摘要 |
PURPOSE:To attain cost reduction and miniaturization of hardware with almost no reduction of sound quality. CONSTITUTION:An input sound signal is separated into a pitch, a PARCOR efficient, a long-term prediction coefficient such as gain, etc., and a long-term prediction residual component except it by a long-term predicting part 2. A recurrence formula data string generating part 7 is preset by various kinds of initial value Is outputted from an initial value sweep part 8 to generate various kinds of recurrence formula data strings which are decided by the initial values Is. A prediction error vector obtained by frame-dividing the long- term prediction residual component and the recurrence formula data string generated by the recurrence formula data string generating part 7 are supplied to a vector distance calculating part 6 and the vector distance of both is calculated. A min. distance detecting part 9 outputs the initial value Is generating the recurrence formula data string which gives min. distance together with the long-term prediction coefficient. |