论文部分内容阅读
汉语是声调语言,相同的音节带上不同的声调所代表的语义就很不相同。为了便于计算机辅助语言学习或用于普通话水平测试系统,准确地检测出声调的发音错误,该文采用精细的上下文相关的声调建模(contextdependent tone model,CDTM),并通过度量与实际发音最相符合的声调模型与预期的声调模型间的KL散度(Kullback-Leibler Divergence,KLD)来检测声调发音的正确性。实验结果表明,在控制错误接受率和错误拒绝率相等的前提下,错误接受率约为6.7%。
Chinese is a tonal language, and the semantics represented by different tones in the same syllable are quite different. In order to facilitate computer-aided language learning or to use the Mandarin level test system to accurately detect pitch phonetic errors, this paper uses a contextual dependent tone model (CDTM) and uses the most accurate measure of actual pronunciation The Kullback-Leibler Divergence (KLD) between the matched tone model and the expected tone model is used to detect the correctness of tone pronunciation. The experimental results show that the error acceptance rate is about 6.7% under the premise of equal control error acceptance rate and false rejection rate.