G.729.1语音编码算法的改进及实现

Improvement and implementation of G.729.1 speech coding algorithm

在G.729.1语音编码算法中引入了一种长度为N=5×2m(m≥2)的MDCT/IMDCT实现的快速算法,编程实现该算法并在cygwin中仿真。在本方法中用孤立预乘的方法将N点的MDCT映射成N/2点的DCT-Ⅱ和DCT-IV,然后进入加窗阶段。该窗口是一种部分对称的窗,可以存储使用N/2的字,同时包含了一种只需4次乘法在一定的条件下就可以分解5点DCT-Ⅱ的方法。与其他算法相比,该算法可以明显地降低复杂度和减少存储空间。

This paper describes an efficient algorithm for implementing MDCT/IMDCT of length N=5×2^m（m≥2） in G.729.1. The algorithm has implemented in float-point on the platform of cygwin. It utilizes a mapping of MDCT of size N into N/2-point DCT-Ⅳ and DCT-Ⅱ with isolated premuhiplications, which are subsequently moved in the windowing stage. The window is piece-wise symmetric, and can be store N/2 words. The algorithm also uses an efficient factorization of 5 point DCT-Ⅱ which requires only 4 multiplications. This algorithm offers practically appreciable reduction in complexity and memory usage.