| 用于语音识别中的SOFM矢量量化方法 |
| |
| 引用本文： | 林宝成,黄志同.用于语音识别中的SOFM矢量量化方法[J].南京理工大学学报(自然科学版),1996,20(1):59-62,74. |
| |
| 作者姓名： | 林宝成 黄志同 |
| |
| 作者单位： | 东南大学无线电工程系,南京理工大学信息学院 |
| |
| 基金项目： | 国家教委博士后资金,江苏省自然科学资金 |
| |
| 摘 要： | 该文讨论了神经网络语音识别系统中的ＫｏｈｏｎｅｎＳＯＦＭ神经网络的矢量量化（ＶＱ）的原理与过程,着重解决了其在实际应用时的若干问题,包括网络大小,学习步幅函数以及矢量量化过程中的公平竞争学习的控制函数的选定。过讨论了基于听觉模型的声学特性的提取与处理,这与ＶＱ性能息息相关。系统应用的实验表明,该神经网络在语音识别系统中具有良好性能。
|
| 关 键 词： | 神经网络 语音识别 自组织特征映射 矢量量化 |
A Method of SOFM as a Vector Quantizer in Speech Recognition |
| |
| Lin Baoeheng,Huing Zhitong.A Method of SOFM as a Vector Quantizer in Speech Recognition[J].Journal of Nanjing University of Science and Technology(Nature Science),1996,20(1):59-62,74. |
| |
| Authors: | Lin Baoeheng Huing Zhitong |
| |
| Abstract: | VQ theory and process of Kohonen SOFM neural network,which is used in our"Speech Recognition based on Neural Network"is presented. Several practical problems such as the choice of net size,learning step function and control function of fair competition learning in VQ process are solved.Acoustic feature based on allditory model is also presented which is important to the performance of VQ. Application of the SOFM VQ neural network to our speech recognition system has shown that it has goodperformance. |
| |
| Keywords: | quantization neural networks speech recognitionl self-organizing feature maps auditory model |
| 本文献已被 CNKI 维普 等数据库收录！ |
|
