摘要
针对传统端点检测算法因忽视语言特点导致的检测精度不足问题,结合元音中心论、响度说、合张运动说等俄语音节理论,提出一种面向俄语语音的音节端点检测算法.算法利用共振峰能量检测元音,并根据元音动态调整门限,基于短时过零率和能熵比提取和切分音节.算法在元音检测中查准率为84.9%,查全率为87%,音节切分的正确率为78.6%,端点检测精度为91.6%,较传统算法剔除了音节间的无话帧,提高了端点检测的精度.
Aiming to solve the problem that the traditional endpoint detection algorithm lacked accuracy due to ignoring the language features, an algorithm of syllable endpoint detection was presented based on Russian syllable theories, such as the vowel center theory, the loudness theory, the motion theory and so on. The formant energy was used to detect vowels, then the thresholds were adjusted according to vowel. Extracts and segments syllables were carried out based on short-time zero crossing rate and energy entropy ratio. The results showed that, the precision was 87% , and the recall rate was 84. 9 % in vowel detection of Russian speech, and the precision was 78. 6 % in syllable segmentation. In addition, the algorithm could improve the accuracy of endpoint detection by eliminating the invalid frames between the syllables.
作者
王彤
易绵竹
WANG Tong;YI Mianzhu(Department of Engineering, Luoyang Branch of Information Engineering University,Luoyang 471000 , China)
出处
《郑州大学学报（理学版）》
CAS
北大核心
2017年第4期34-39,共6页
Journal of Zhengzhou University:Natural Science Edition
基金
国家自然科学基金项目(11590771)
关键词
元音检测
俄语语音音节切分
端点检测
vowel detection
Russian syllable segmentation
endpoint detection
