摘要
目前基于纯净语音信号的语音识别系统和说话人识别系统都已达到了很高的识别率,但是当信号中含有噪声,特别是含有语音噪声时,识别率就会大大降低.解决这一问题的关键是实现语音与噪声的自动分离.考虑到语音信号的非平稳特性,把时域去相关的思想推广到频域,提出了频域去相关算法,实验结果显示了算法的有效性.
The recognition system of speech signals and speakers based on pure speech signals has a good performance, but the performance of the system will be decreased when there are noises in the signals. The key to solving this problem is automatic separation of speech and noise.Decorrelation in time domain is extended into frequency domain and given an algorithm of decorrelation in the frequency domain is given based on the non_stationary of speech signals. Simulation results show that this method is effective.
出处
《应用科技》
CAS
2005年第2期53-55,共3页
Applied Science and Technology
关键词
短时傅立叶变换
频域
去相关
语音信号分离
short-time Fourier transform
frequency domain
decorrelation
speech separation