Skip to Main Content
In order to obtain a high quality in speech synthesis, it is necessary to increase the frame rate in speech analysis. It has not been thoroughly investigated how frequent the optimal frame rate should be. First, we examined time fluctuating characteristics of the LPC parameters, such as log area ratio (LAR) and line spectrum pair (LSP) by spectral analysis. Next, we discuss the spectral interpolation distortion of these parameters, and the optimum coding to minimize the distortion. In case of LAR, it is possible to decrease to decrease the frame rate by 10 % through a non-uniform allocation of frame rates among different parameters. In case of LSP, the non-uniform allocation of frame rates is not necessary because the fluctuation characteristics of spectral interpolation distortion do not heavily depend on the order of LSP.