A neural network method for generrating the linear prediction coefficients of random sequences 一種求線性預(yù)測系數(shù)的神經(jīng)網(wǎng)絡(luò)方法
The adaptation processing includes linear prediction coefficient adaptation and adaptation of quantization step size for residual signals . based on g . 726 , we adopt a huffman coder to make use of probability statistic of bit cascade covering n ( n 1 ) samples generated from adpcm , in order to further reduce the bit rate . ng is lossless entropy coding , the speech quality of our improved algorithm should be same as that of g . 726 standard 我們的研究和改進工作包括:研究最優(yōu)非均勻自適應(yīng)量化器,及其自適應(yīng)算法;研究波形預(yù)測函數(shù),以及函數(shù)零點、極點的自適應(yīng)算法;基于每n ( n 1 )個樣本所對應(yīng)符號的概率統(tǒng)計,對預(yù)測殘差量化值再進行huffman編碼,進一步降低比特率。