摘要
提出了一种新的文本无关的说话人识别方法,它通过KL变换对语音进行规整、去噪,再利用小波包分解系数弹性地选择频带,提高时频分辨率,更好地提取出说话人的特征。实验证明,这种方法不仅具有较高的识别率,并且在嘈杂环境下也能保持稳定的性能。
A new method for text-independent speaker recognition is proposed.It can extract speakers' feature commendably by making use of wavelet packet decomposition coefficients to choose frequency band elastically and raise time-frequency resolution,after warping and denoise speech with KL transform.Experiment results show that this method can not only achieve high recognition rate,but also remain stable under noisy environment.
出处
《计算机工程与应用》
CSCD
北大核心
2005年第4期26-28,共3页
Computer Engineering and Applications
基金
国家自然科学基金资助项目(编号:60135010)
关键词
说话人识别
KL变换
小波包分析
speaker recognition,KL transform,wavelet packet analysis
