Development of Foreign Accent Identification System and Building of Non-native Japanese Speech Corpus

该项目的目的是建立一个日语母语和非母语语料库，并开发一个外国口音识别系统。在语音语料库中，33名母语人士和45名非母语人士通过固定电话和麦克风录制了包括单词、句子、数字、段落在内的各种日语话语。然后将录制的语音材料提交声学分析，以找出母语和非母语语音之间的差异。利用反映非母语人士语音特征的语音号码的韵律和阅读段落的发音速率进行外国口音识别实验，结果表明我们可以获得更高的识别准确率。

The purposes of this project were to build a corpus of native and non-native Japanese speech and to develop a foreign accent identification system. For the speech corpus, thirty-three native and forty-five non-native speakers recorded various Japanese utterances including words, sentences, numbers, and passages through the landline and a microphone. The recorded speech materials were then submitted to the acoustic analyses in order to find the differences between native and non-native speech. Foreign accent identification experiments were conducted using the prosody of spoken telephone numbers and the articulation rate of read passages, which reflected the acoustic characteristics of non-native speakers, and the results showed that we could obtain higher identification accuracy.