针对口语语音处理中的基频提取,提出了一种改进的自相关函数基频提取算法。该算法在原始自相关函数方法的基础上,通过利用语音频谱的纹理特征来提高正确基频值的权重,利用增加候选基频的个数来增大搜索空间,以及利用可靠种子来限制搜索路径这3项措施增加了正确基频值在搜索空间中的出现比例和权重,优化了搜索空间,从而改善了原有基频提取算法的性能。在数据集Keele和FDA上的实验结果显示:与原始算法相比,本文算法的有声错误率相对减少28.74%,总体错误率相对减少5.53%,更适合于口语处理。 Aiming at the fundamental frequency extraction in spoken speech processing, an improved basic frequency extraction algorithm for autocorrelation function is proposed. Based on the original autocorrelation function method, this algorithm increases the weight of the correct base frequency by using the texture features of the speech spectrum, increases the search space by increasing the number of candidate fundamental frequencies, and uses the reliable seed to limit the search path These three measures increase the proportion and weight of the correct fundamental frequency in the search space, and optimize the search space, so as to improve the performance of the original fundamental frequency extraction algorithm. Experimental results on data set Keele and FDA show that compared with the original algorithm, the proposed algorithm has a 28.74% reduction in the acoustic error rate and a 5.53% reduction in the overall error rate, which is more suitable for spoken language processing.