摘要
中国地域辽阔,汉语包括多种方言。方言转换技术可实现方言间语音的转换,丰富语音交流的方式,增强语音交流的可懂性、友好性和趣味性。该文分析了普通话与沈阳方言在声调调类与调值、时长和特殊词等方面的差异;探讨了普通话与沈阳方言转换的可行性;提出了基于混合聚类算法的方言基频转换模型、基于统计的方言时长转换模型和基于加权有限状态机的特殊词识别算法;通过基本转换和说话人音色还原构建了方言转换系统,实现了实时普通话语音输入、沈阳方言语音输出。评测结果显示通过方言转换系统得到的方言语音与目标方言语音具有较高相似度。
China has a variety of dialects so dialect conversion systems will facilitate communication between people who speak different dialects.This paper analyzed the differences between Mandarin Chinese and the Shenyang dialect in terms of pitch contour,duration and special words.A pitch contour conversion model was then developed based on a mixed clustering algorithm using agglomeration clustering and k-means clustering methods a statistical duration conversion model and a continuous speech spotting algorithm for the special words.Tests of the dialect conversion system for Mandarin Chinese to the Shenyang dialect show a high similarity between the original Shenyang dialect sentences and converted sentences from Mandarin Chinese.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2009年第S1期1309-1315,共7页
Journal of Tsinghua University(Science and Technology)
基金
国家自然科学基金资助项目(60805008
90820304)
国家"八六三"高技术项目(2007AA01Z198)
关键词
方言转换
基频聚类
时长统计
语音识别
音色还原
dialect conversion
pitch contour clustering
duration statistics
speech recognition
pitch contour adjusting
