欢迎您！东篱公司
退出
-
申报数据库
-
立项数据库
-
成果数据库
-
项目获奖数据库
中文摘要：
针对语音多变性的鲁棒性问题,分别将FIR滤波器、Gammatone（GT）滤波器、Laguerre滤波器以及弯折滤波器（Warped Filter Banks,WFBs）用于过零峰值幅度（Zero Crossing Peak Amplitude,ZCPA）特征提取,并使用支持向量机（Support Vector Machine,SVM）作为后端识别系统,通过实验得到了不同滤波器下ZCPA的识别结果。结果表明在多变性语音的识别中,SVM系统较常用的HMM系统,更适于ZCPA特征;并且在SVM系统下,ERB尺度的弯折滤波器较其它前端滤波器识别效果更好,明显优于常用的MFCC特征。
英文摘要：
In order to solve the variability robustness in speech recognition systems,different front-end filter banks such as FIR filter bank,Gammatone（GT） filter bank,Laguerre filter bank and Warped Filter Banks（WFBs） were used to extract Zero Crossing Peak Amplitude（ZCPA） feature.They were all based on the Support Vector Machine（SVM） system.The experiments show that the SVM was much more suitable for ZCPA than HMM in variability recognition tasks.Moreover,in the SVM system,the ERB-scale WFBs had the best recognition results compared with the other front-end filter banks.It outperformed significantly than MFCC.
同期刊论文项目
同项目期刊论文
Parameter optimization and application of support vector machine based on parallel artificial fish s
Study for classification of emotional speech by using optimized frame zero crossing with peak amplit
A robust feature extraction approach based on an auditory model for classification of speech and exp
期刊信息
位置：