基于改进神经网络的三字词声调识别方法  

Tone Recognition of Three-syllabic Words Based on Improved Neural Network Algorithm

作　　者：郑燕琳[1] 许星宇[1] 杨晓炯[1] 

机构地区：[1]公安部第三研究所,上海200031

出　　处：《电声技术》2010年第5期48-51,共4页Audio Engineering

摘　　要：汉语连续语音三字词声调由于受声道响应、音连、变调规律等的影响,情况较复杂,声调模式间交集大,模糊性强。提出了一种基于动态时间规整和改进神经网络的声调识别算法。实验表明,该算法可得到较好的识别效果,对探讨具有汉语特色的新的语音识别方法有重要意义。The continuous Chinese tone of three-syllabic word is very complicated and its tone patterns lie in illegibility because of the effects of vocal response, syllable conjunction and tone sandhi. A method of tone recognition based on dynamic time alignment and improved neural network algorithm is presented. The simulation experiments show that the method can get a good recognition result. The method has an important meaning in exploring new speech recognition methods that possess Chinese characters.

关 键 词：连续语音三字词 声调识别 动态时间规整 改进神经网络算法 

分 类 号：TN912[电子电信—通信与信息系统]