| 基于GARCH模型的改进的倒谱域特征参数补偿算法 |
| |
| 引用本文： | 简志华,杨震.基于GARCH模型的改进的倒谱域特征参数补偿算法[J].信号处理,2007,23(3):383-387. |
| |
| 作者姓名： | 简志华 杨震 |
| |
| 作者单位： | 南京邮电大学信号与信息处理研究所,江苏南京,210003 |
| |
| 基金项目： | 江苏省教育厅青蓝工程项目 |
| |
| 摘 要： | 本文提出了一种改进的倒谱域特征参数补偿算法GMCSM。根据语音信号的时变特性,GMCSM算法使用广义自回归条件异方差(Generalized Auto-Regressive Conditional Heteroscedasticity,GARCH)模型对语音信号的方差进行建模。实验数据表明,与常规倒谱相减法CSM和MEMCSM相比,GMCSM能够更有效地补偿因加性噪声引起的倒谱特征参数失真,减少识别的错误率,特别是在信噪比较低的情况下,GMCSM的性能更为显著。
|
| 关 键 词： | 语音识别 鲁棒性 广义自回归条件异方差模型 |
| 修稿时间： | 2005年11月29 |
A Modified Method of Cepstrum-Domain Feature Compensation Based on GARCH Model for Robust Speech Recognition |
| |
| Jian Zhihua,Yang Zhen.A Modified Method of Cepstrum-Domain Feature Compensation Based on GARCH Model for Robust Speech Recognition[J].Signal Processing,2007,23(3):383-387. |
| |
| Authors: | Jian Zhihua Yang Zhen |
| |
| Abstract: | GMCSM,a modified feature compensation algorithm in cepstrum domain,is proposed in this paper.In this algorithm, the variance of speech signal is modeled as a GARCH stochastic process according to the time-variant properties of speech signal.The experimental results demonstrate that GMCSM can achieve better performance than CSM and MEMCSM on compensating the distortion of cepstral feature due to additive noise and reducing the word error rate (WEB)of speech recognition,especially under lower SNR condition. |
| |
| Keywords: | GMCSM |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
|
