摘要
针对梅尔倒谱系数与翻转梅尔倒谱系数在语种识别应用中的不足,采用高斯滤波器代替三角滤波器,提出一种新的梅尔倒谱系数提取方法,解决传统梅尔倒谱系数提取中邻近滤波器相关性较弱的问题,并结合Fisher准则构造出最优混合特征参数,采用高斯混合模型分别对不同混合特征进行语种识别。实验结果表明:基于高斯滤波器及Fisher准则的改进梅尔倒谱系数混合特征参数作为语种识别特征具有较高的识别准确率。
For the shortcoming of MFCC and IMFCC used in language identification,this paper uses Gaussian Filter instead of Triangular Filter,and proposes a novel MFCC feature sets extracting method to solve the weak correlation of adjacent filter.Optimistic hybrid parameter is constructed through the combination of GMFCC and GIMFCC using Fisher criterion.Gaussian mixture model is adopted in language recognition with this novel hybrid parameter.Experimental results show that the improved MFCC hybrid parameter based on Gaussian Filter and Fisher criterion is effective in language recognition.
出处
《电路与系统学报》
北大核心
2013年第2期400-404,共5页
Journal of Circuits and Systems
基金
国家自然科学基金资助项目(61272333)
