| 摘要 |
PURPOSE:To provide the speech encoding method which improves the power of expression of the pitch sound source vector of an input speech having a different pitch period without increasing the amount of information. CONSTITUTION:The input speech 9 is analyzed by a long-period pitch analyzing means 3 and a short-period pitch analyzing means 4 to detect a long-period pitch 12 and a short-period pitch 13, and prediction residues matching the pitches are selected from a prediction residue storage means 2 to generate a long-period pitch sound source vector 14 and a short-period pitch sound source vector 15, which are put together to generate a pitch sound source vector 16. A synthesized speech 17 is generated by using the pitch sound source vector 16 and fed back to the pitch analyzing means so that the difference from the input speech 9 is minimized by an error minimizing means 8; and a pitch sound source vector corresponding to the input speech 9 having the different pitch period is generated and the frequency of the long-period pitch analysis and the frequency of the short-period pitch analysis are set to mutually different values in a speech frame, so that the information quantity is prevented from increasing. |
