【摘要】 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。更多还原

【Abstract】 Chinese is a language with syllabic tone. Tone information is very important to continuous speech recognition and synthesis. In this paper, the proposed syllabic tone recognition method extracts features of every syllable in trisyllabic word respectively, chooses corresponding membership function based on the tone characteristic of Chinese trisyllabic word. Improved back-propagation network is applied to syllabic tone recognition and tone recognition for Chinese trisyllabic word, the recognition rate is high. This method is significative to make use of tone information in continuous speech recognition. Furthermore, a novel architecture of back-propagation network is proposed in this paper.更多还原