Vocal tract parame in the audio coded and decoding vessel - be adapted compressed manner of ta information

摘要

PURPOSE:To reduce the deterioration of a speech signal due to digitization as much as possible by adaptively inputting individual speech features of a speech to the encoder and decoder as to speech encoding and decoding technology. CONSTITUTION:When the volcal track parameter is compressed for speech encoding, two neural networks 25 and 26 are used; and one load fixed type neural network 24 performs a mean weighting process and the other learning type neural network 26 learns on the basis of an input parameter to the neural network. In the beginning of the speech encoding, the load fixed type neural network which performs the mean weighting process is used to detect a section which exerts no influence on a conversation and then switched to the learning type neural network 26 which has learnt weight.