面向拉萨语的自动发音错误检测方法研究

Automatic pronunciation error detection is to make the computer point out the concrete errors in the pronunciation of learners which is a interdisciplinary subject. There are three main dialects in Tibetan language,Lhasa, Khamba and Amdo and the Tibetans of different regions find it hard to communicate with others. As the times goes on, Lhasa dialect has become the standard language in Tibetan language with its users more and more. The automatic pronunciation error detection technology for Lhasa dialect has been out of place and the key technology is still in research. This project aims to study the pronunciation error detection technology based on statistical pattern recognition, analyze the regular pattern in the wrong pronunciations, find the strategies in automatically detecting the pronunciation errors in line with the pronunciation patterns of Lhasa dialect by prior probability algorithm in wrong pronunciations, differential acoustic characteristics extraction methods, the pronunciation space model based on speech evaluation, and adaptive algorithm in automatic detection of pronunciation error, form the technological solutions and stimulates Tibetan speech evaluation project.

自动发音错误检测就是让计算机像人一样指出学习者发音中的具体发音错误，是一门典型的交叉学科。藏语存在拉萨、康巴、安多三大方言，各地藏族同胞的交流不便。随着时代的发展，拉萨语已逐渐成为藏语的"普通话"，学习使用者越来越多。拉萨语的自动发音错误检测技术目前处于空白，其关键技术亟待解决。本项目以基于统计模式识别的发音错误检测技术为研究目标，通过模型分析拉萨语的发音错误规律，利用发音错误的先验概率算法、区分性声学特征提取方法、针对语音评测的发音空间模型方法、发音错误自动检测的自适应算法等技术，研究符合拉萨语发音规律的自动发音错误检测策略，形成技术解决方案，从而促进藏语语音评测工程化步伐。

自动发音错误检测就是计算机像人一样指出学习者发音中的具体发音错误，是一门典型的交叉学科。藏语存在拉萨、康巴、安多三大方言，各地藏族同胞的交流不便。随着时代的发展，拉萨语已逐渐成为藏语的“普通话”，学习使用者越来越多。拉萨语的自动发音错误检测技术目前处于空白，其关键技术亟待解决。本项目经过一年的研究，完成了藏语语音数据采集、数据标注、建立了藏语语音评测数据库、找出了藏语拉萨语的发音规律，为下一步研发藏语语音评测系统完善了基础资源。