摘要
为解决语音识别过程中的抗噪声及抗干扰问题,提高系统的识别精度,利用隐马尔可夫模型HMM优异的时序建模能力及小波变换可以对信号进行多尺度分析并有效提取信号的局部信息的特点,建立了混合语音识别模型.考虑到在语音信号识别过程中信号的非平稳性,采用并行的识别方法分别获取分类信息,根据混合模型的识别算法做出识别决策,减小了系统对环境的依赖性,提高了其自适应能力.仿真实验结果表明,混合模型识别结果比单一HMM模型或小波模型识别结果更佳,提高了整体的识别速度和识别率.
To solve anti- noise and interference problems in the speech recognition process and improve the recognition accuracy, in this article, dynamic time sequence modeling of hidden markov model （HMM） and wavelet analysis are applied to extract more effective the local information of signals, and set up a hybrid speech recognition model. In the process of voice signal identification, considering nonstationarity of phonetic signal, parallel identification methods are used to obtain classified information. The result of recognition is made by using recognition algorithm of the hybrid model, it reduces the system＇s dependence on the environment and improves its adaptive capacity. Recognition experiment shows that this hybrid model has higher performance than hidden Markov model in noisy speech recognition.
出处
《河南理工大学学报（自然科学版）》
CAS
2007年第6期694-699,共6页
Journal of Henan Polytechnic University（Natural Science）
基金
国家自然科学基金资助项目(604740437)
作者简介
张丽（1982-），女，山西临汾人，研究方向为智能控制与信息处理技术E-mail：zhangli00121@163．com
