摘要
针对时域同步叠加时长规整算法计算量大,不适合在实时语音处理中应用的情形,提出了一种新的基于同步叠加方法的时长规整算法——自适应同步叠加算法,根据语音信号的准周期性质,即时调整搜索算法中的搜索间隔,以最快的速度找到最准确的叠加位置,从而达到实时处理的目的。该算法的分析测试表明,合成的语音质量高、实时性好,可以有效地应用到实时语音处理场合中。
Synchronous overlap and add (SOLA) method was popularly used for time scale modification, but this method had high computational cost. In this paper, a novel time scale modification algorithm named adaptive synchronous overlap and add (ASOLA) was presented. Based on the pre-period property of voice speech, the searching interval was modified timely so that the best overlap point is sought at the fastest speed, and time scale modification for speech is processed in real-time. Experiment results show that ASOLA has low computational cost, the speech synthesized by ASOLA has high quality, and ASOLA can be effectively used in real-time speech processing.
出处
《通信学报》
EI
CSCD
北大核心
2005年第2期136-140,共5页
Journal on Communications
基金
国家自然科学基金资助项目(60273040)
江苏省高校自然科学基金资助项目(02KJB520003)
关键词
语音处理
时长规整
同步叠加
实时
speech processing
time-scale
synchronous overlap and add
real-time