论文部分内容阅读
汉语是音节声调语言,在连续语音合成和识别中加入声调信息有重要意义。本文对三字词声调按首字、中字和尾字音节进行特征提取,并结合声调的特点选取相应的隶属函数进行模糊化处理,结合神经网络技术进行音节声调和三字词声调识别,取得了较好的实验结果。该方法对连续语音识别中利用声调信息有参考价值。此外文中还给出了一种新的BP网络结构。
Chinese is a syllable tonal language. It is of great significance to add tonal information to continuous speech synthesis and recognition. In this paper, feature extraction is made on the first word, the middle word and the tail syllable based on the first word, the middle word and the tail syllable, and the corresponding membership functions are selected according to the characteristics of the tone to be fuzzified. The syllabic tone and three-word tone identification are combined with the neural network technology, Better experimental results. The method has reference value for using tone information in continuous speech recognition. In addition, the article also gives a new BP network structure.