摘要
藏语数字语音识别是藏语语音识别中的基础环节。文章以藏语数字语音信号产生的数学模型为研究对象,从频域、时域、倒谱域出发,对藏语数字语音识别进行了分析,重点论述了藏语数字语音识别的原理、仿真实验及测试方法。以藏语数字语音库为数据资源,对藏语数字语音建立了训练模型并从中提取语音特征,通过模板匹配算法,获取录音数据中待识别语音的模板。实验证明已建立语音库的识别率达100%,随机录音的识别率则达90%。
Tibetan digital voice recognition is the baseline procedure in Tibetan voice recognition. In this paper,the mathematical model of Tibetan digital voice signal generation as the research object, Tibetan digital voicerecognition was analyzed based on the frequency domain, time domain, cestrum domain, and mainly discoursedthe theory, emulation experiment and testing method of Tibetan digital voice recognition. A training model for Ti-betan digital voice was established to extract the voice features in it based on Tibetan digital voice database andthe voice to be recognized in the recorded data was obtained by template matching algorithm. The experimentshowed that recognition rate of the established voice database was up to 100% and that of random recording wasup to 90%.
基金
2014年度西藏自治区一般科技计划项目"基于小词汇量的拉萨语声学特征提取研究"(项目号:2015ZR-14-1)
2014年度西藏大学青年科研培育基金项目"面向拉萨语的语音声学特征提取研究"(项目号:ZDPJZK201403)
2014年度国家哲学社会科学基金重大项目"基于地理信息平台的藏语方言数据库建设"(项目号:14ZDB101)阶段性成果
关键词
藏语语音识别
仿真实验
语音测试
Tibetan voice recognition
emulation experiment
voice test
