| 摘要 |
PURPOSE:To encode a voice with high quality by introducing constitution which approximates variation in the linear predictive parameter in a frame and using the approximated linear predictive parameter to find a sound source parameter. CONSTITUTION:A voice signal is stored in an input buffer 110 by N samples for one frame through an input terminal 100. A linear predictive analyzer 120 extracts linear predictive parameters from the input speech signal at intervals of a subframe shorter than the frame length, performs representation and quantization with the extracted linear predictive parameters at the subframe intervals, and then outputs the results to an auto-correlation function calculator 130, a cross-correlation function calculator 140, and a linear predictive encoder 170. A pulse searcher 150 extracts sound source parameters from the calculated autocorrelation function and cross-correlation function. The sound source parameter and linear predictive parameter are encoded, multiplexed by a multiplexer 180, and then sent to a decoding side through an encoding output terminal 190. |
