论文部分内容阅读
汉语是一种带声调的语言,声调信息在汉语识别中具有非常重要的意义.传统的声调识别一般只研究朗读式语音中相对标准的声调,很少对声调调型比较复杂的自然口语进行专门的处理.针对汉语自然口语的特点,在声调建模单元的选择时提出了真实上下文的模型.同时,为了对声调模式进行精细建模,采用了一种层次聚类的方法来获得更多的声调模式.实验结果证明了方法的有效性.
Chinese is a tonal language, and tonal information is of great importance in Chinese recognition.Traditional tonal recognition generally only studies the relative standard tones in spoken-word speech, and seldom studies the more complex natural spoken words in tone-tuning Aiming at the characteristics of Chinese natural spoken language, a real context model is proposed in the selection of tone modeling unit.At the same time, in order to finely model the tone mode, a hierarchical clustering method is adopted to obtain more Tone mode.The experimental results prove the effectiveness of the method.