摘要
该文提出了一种将模糊C-均值聚类法的各种改进算法与矢量量化法相结合的说话人辨认的新方法。首先从语音信号中提取MFCC特征矢量,其次利用矢量量化来设计码书,最后用改进算法对待识语音进行辨认。新算法的辨认率达到95%以上,抗噪性能也优于矢量量化法。
Several new algorithms of fuzzy C-mean clustering with the combination of vector quantization are proposed for speaker identification.First the Mel -frequency Cepstral Coefficients are extracted from speech signals.Second,codebooks are designed using vector quantization approach.At last,someone's speeches are identified using the new algorithms of FCM.It is proved that identification rate of the algorithms is more than95%and robust of them is super to vector quantization approach.
出处
《计算机工程与应用》
CSCD
北大核心
2003年第27期94-95,140,共3页
Computer Engineering and Applications
关键词
模糊C-均值聚类法
矢量量化
模拟退火算法
遗传算法
进化免疫算法
Fuzzy C-mean clustering,Vector quantization,Simulated annealing algorithm,Genetic algorithm,Immune evolu-tionary algorithm