Abstract: A low complexity fractional pitch estimation algorithm for speech coding based on polynomial fitting theory is presented.Giving a series of long term prediction gain at integer sampling points，the fractional pitch is estimated by quadratic fitting at the proximity of the maximum point.Compared to the known algorithm based on sampling function interpolation，it eliminates constant interpolation table，interpolations，exhaustive searching，and comparison operations，with a complexity independent of interpolation factor—1/25 of that of the known algorithm in typical cases.The objective and subjective tests of a speech coder using the proposed algorithm show equivalent performance in terms of long term prediction gain and speech quality as the known algorithm.

Key words: speech coding, fractional pitch, pitch estimation, interpolation, polynomial fitting

摘要： 在保证同等音质的前提下，为降低语音编码器中分数基音估计的复杂度，提出一种基于多项式拟合的分数基音估计算法。以整数点相关度序列为基础，在其最大值点附近进行多项式拟合，解析给出分数基音估计值。与现有的基于采样函数插值的分数基音估计算法相比，不使用常数插值表，无插值、遍历和比较操作，实现运算量与插值因子无关，典型情况下的运算复杂度仅为现有算法的1/25。在实际语音编码器中，所提算法对各类语音的客观及主观测试结果表明，其长时预测增益和编码音质都与现有算法相当。

关键词: 语音编码, 分数基音, 基音估计, 插值法, 多项式拟合