摘要
使用独立分量分析(ICA)来提取说话人特征并与矢量量化(VQ)判决方法相结合,实现了一个高性能的基于ICA特征的VQ(ICA-VQ)说话人识别系统。通过ICA变换得到说话人语音特征基函数系数用于生成VQ码书,并导出包含能量失真的ICA-VQ码书失真测度和质心确定条件,生成最终的判决。仿真实验中ICA提取的特征分别用于不同系统实现说话人确认任务,各系统的DET曲线对比验证了VQ方法用于ICA特征分类判决的优势,同时不同码书尺寸下的等差率(EER)对比证明了VQ码书设计的有效性。
The paper combined the speaker feature extracted by ICA with VQ technique to the ICA-VQ speaker recognition system with high performance. A speaker speech ICA synthesis model was presented to get the speaker speech feature bases with ICA algorithm, and the coefficients of the bases were used in designing codebooks. A novel distortion measurement including energy and a new centroid condition were given. In the simulation experiment of speaker verification, the EER contrast results of VQ with different sizes prove that VQ codebooks are efficient and the DET cures of various methods show that VQ is a more suitable method to speaker recognition with the coefficients of ICA feature bases.
出处
《计算机应用》
CSCD
北大核心
2005年第10期2401-2403,共3页
journal of Computer Applications
基金
国家十五科技攻关课题(2004BA616A1103)
关键词
独立分量分析(ICA)
矢量量化(VQ)
说话人识别
失真测度
Independent Component Analysis (ICA)
Vector Quantization (VQ)
speaker recognition
distortion measurement