论文部分内容阅读
基音周期是语音信号的一个重要参数。为了在低延时条件下准确地提取语音基音周期参数,提出了一种基于动态规划的单路径搜索算法。该算法引入了清浊音信息和帧间相对能量对相邻帧代价函数进行加权;使用了一种更加有伸缩性的方式对基音变化幅度进行控制,以代替基音变化的硬性限制;专门针对基音突变和自身错误的情况,对过去帧影响进行限制,在保持基音跟踪性能和允许基音突变之间寻找一个平衡点。最后,使用Keele数据库进行的测试表明,该算法在只有一帧延时的情况下,严重错误率比传统算法下降2.32%。
The pitch period is an important parameter of the speech signal. In order to accurately extract speech pitch parameters under low delay conditions, a single-path search algorithm based on dynamic programming is proposed. The algorithm introduces the voiced sound information and the inter-frame relative energy to weight the adjacent frame cost function. It uses a more scalable way to control the amplitude of the pitch change, in order to replace the hard limit of the pitch change. And their own mistakes, limiting the impact of past frames, looking for a balance between maintaining pitch-tracking performance and allowing pitch bursts. Finally, tests using Keele database show that the proposed algorithm reduces the critical error rate by 2.32% compared with the traditional algorithm with only one frame delay.