| 基于多带HMM和神经网络融合的语音识别方法的信道鲁棒性 |
| |
| 引用本文： | 姚志强,戴蓓倩,李辉,黄伟.基于多带HMM和神经网络融合的语音识别方法的信道鲁棒性[J].计算机工程与应用,2004,40(1):71-73,82. |
| |
| 作者姓名： | 姚志强 戴蓓倩 李辉 黄伟 |
| |
| 作者单位： | 中国科学技术大学电子科学与技术系,合肥,230026 |
| |
| 基金项目： | 安徽省自然科学基金项目资助(编号:01042205) |
| |
| 摘 要： | 对于基于HMM的语音识别系统,由于训练和测试环境(背景噪声。语音传输信道Microphone等)的失配将会造成识别性能的严重下降。根据人类的听觉感知机理,该文针对语音传输信道失配问题,提出了一种基于多带HMM的系统结构,有若干个子带系统和一个全频带子系统组成,并采用神经网络对个子系统的输出进行后端融合及判决。实验表明,这种方法可以有效地提高识别系统的信道鲁棒性。
|
| 关 键 词： | 信道失配 子带HMM 神经网络融合 信道鲁棒性 |
| 文章编号： | 1002-8331-(2004)01-0071-03 |
Channel Robust of Speech Recognition Based on Multi-band HMM and BPNN Fusion |
| |
| Yao Zhiqiang Dai Beiqian Li Hui Huang Wei.Channel Robust of Speech Recognition Based on Multi-band HMM and BPNN Fusion[J].Computer Engineering and Applications,2004,40(1):71-73,82. |
| |
| Authors: | Yao Zhiqiang Dai Beiqian Li Hui Huang Wei |
| |
| Abstract: | Due to the mismatch between trai ni ng and testing environment (such as background noise,speech transition channel-Microphone),automatic speech recognition(ASR)based on HMM tends to drastic ally degrade in performance.For the reason of the theory of human aural percept ion,in this paper,we present a system based on multi-band HMM,composed of several sub-band systems and a full-band system.Then,we use ANN to make the final decision by combined all systems' outputs.Through test,our system i s proved to be efficient to upgrade the performane of channel robustness of the recognition system. |
| |
| Keywords: | channel mismatch sub-band HMM ANN fusion cha nnel robustness |
| 本文献已被 CNKI 维普 万方数据 等数据库收录！ |
