摘要
给出一种采用LPCCEP特征参数和改进矢量量化的数字音说话人识别方法 ,该方法适于说话人集合适中(10 0人以内 )情况 ,具有训练简单、存储数据量少、响应快等特点 .为提高系统的识别率和可靠性 ,系统采用了方差修正的VQ改进算法 ,使建立的说话人模型考虑了说话人帧特征概率分布的更精细结构 .实验结果表明 。
A method of digits text speaker identification is introduced, which is based on LPCCEP parameters and improved vector quantization modeling. It is suited for a small speaker set (e. g. less than 100 people) and has the advantages of simple training, small storage data and quick response. In order to get higher recognition rate and reliability, the VQ matching model is modified based on the variance so that more information about the speaker identity is taken into the improved model. The good identification effect of the improved method has been confirmed by experiments.
出处
《山东大学学报（工学版）》
CAS
2004年第3期60-62,共3页
Journal of Shandong University（Engineering Science）
关键词
说话人识别
LPCCEP
矢量量化
speaker identification
LPCCEP
vector quantization (VQ)
