一种降低码激励线性预测编码器运算量的方法

A method of reducing computational complexity in CELP coder

降低CELP(码激励线性预测)编码器运算量是语声编码器领域的一个研究热点。本文研究了一种用DFT(离散傅里叶变换)简化CELP编码中最佳激励选择的搜索过程,即利用变换域法简化卷积运算。在计算机模拟实验中对女声(8kHz采样,12bit量化的语音)进行了10s的试听。结果表明:该方法使CELP编码器的计算量减少近10倍,而合成语声质量仍无感性降低。

In speech coding environment,the study is being focused on reducing computational complexity of GELP(Code-Excited Linear Prediction) . An approach is made to simplify the search procedures for optimum-excitation-selection in CELP coder by DFT (Discrete Fourier Transform); i. e. by making use of the transform domain method to simplify convolution operation. By this method, in the period of 10-second audile test for female vice (8kHz sample 12 bit quantization speech sounds) on computer analogue experiment, it enables computational reduction of CELP coder attaining approximately to tenfold; and there is no perceptual degradation of synthetic speech quality.