摘要
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%。
Pitch is one of the most important parameters for describing speech characteristics. A one-way pitch tracking algorithm based on dynamic programming was developed to extract the correct pitch within a limited delay. The Unvolced/Voiced （U/V） and power variety information was utilized to weight the penalty of consecutive speech segments. A flexible method was used instead of hard constraints, to keep the pitch contour continuity. The pitch variation penalty between the previous frame and the current one was restricted to attenuate the influence of the previous frame, since abrupt change in the pitch contour occasionally occur and the algorithm sometimes makes mistakes. Simulations based on the Keele database show that the gross error reduced by 2.32% compared to the traditional algorithm with only a one frame delay.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2008年第10期1586-1588,共3页
Journal of Tsinghua University(Science and Technology)
基金
国家自然科学基金资助项目(60572081)
