摘要
为解决语音处理上对大量具体语料的依赖及其繁重处理的问题,该文首先通过对语音处理的根本目标与语音技术的当前现状的分析,指出了这种依赖性的根源。接着通过对语音多变的不可避免性与声学不变量的相对性的阐述,说明语音的变化并非完全不可知,进而指出解决问题的关键在于充分认识语音变化的规律性和在处理系统中综合利用这些规律。最后,提出一个解决策略,基本原则就是通过完善语料库建设来促进知识与语料的有机结合,逐步以相对关系上的声学不变量来取代具体语料的作用。并对相关语料库的建设提出了初步设想。
This study provides solutions to speech processing system problems when systems rely heavily on large amounts of speech materials and need large amount of computations to deal with these materials.The paper describes the origin of this dependence by analyzing the essential goal and current technologies of speech processing,then explains that analysis of speech variations is not an intractable problem through analysis of speech terms and their relational acoustic invariance.The study further indicates that the key solution is to accurately describe for speech variations and integrate them into the speech processing.The solution promotes the integration of phonetic rules and speech materials by improving the construction of the speech database to introduce more relational acoustic invariance to gradually replace the role of specific speech materials.The construction of the speech database is also provided.
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1380-1387,共8页
Journal of Tsinghua University(Science and Technology)
基金
中国社会科学院老年科研基金项目
中国社会科学院语言所与东芝(中国)研究中心合作项目
关键词
语音多变性
声学不变量
语料库建设
variability of speech sound
acoustic invariance
speech database construction