论文部分内容阅读
本文提出一种用隐马氏链模型识别汉语声调的新方案。由每一种声调的训练语音求出相应的概率模型参数作为识别模板。识别时,分别用每一种声调的模型参数计算出现输入语声周期序列的概率,概率最大者即为输入语声的声调模型。实验语音选用的是“小学汉语拼音教学录音磁带”,一个男声和一个女声,对于其中的24个韵母和21组拼音音节,正确识别率为98%。
In this paper, a new scheme of recognizing Chinese tones by using hidden Markov chain model is proposed. The corresponding probabilistic model parameters are obtained from the training speech of each tone as the recognition template. When recognizing, the probability of the input speech period sequence is calculated by the model parameters of each tone separately. The highest probability is the tone model of the input speech. The experimental phonetic transcription used “Primary Chinese Pinyin Teaching Tape”, a male voice and a female voice. The correct recognition rate was 98% for 24 vowels and 21 phonetic syllables.