【摘要】 汉语方言辨识中常用的转移差分倒谱(SDC)特征往往存在较多的冗余信息.对此,提出动态时频倒谱(DT-FC)特征.首先对倒谱矩阵进行离散余弦变换(DCT),然后对变换后的矩阵元素进行重组.基于新特征,在高斯混合模型系统下对闽、粤、吴3种方言进行辨识.实验结果表明,DTFC特征的性能明显优于SDC特征,其平均辨识率可达98.89%,较SDC特征提高了3.1%.更多还原

【Abstract】 The shifted delta cepstrum(SDC) feature used widely for Chinese dialect identification contains much redundancy.In this paper,a dynamic time-frequency cepstral(DTFC) feature is proposed,which is obtained by performing a discrete cosine transform(DCT) on the cepstrum matrix and rearranging the transformed elements.Experiment results on the Chinese dialects(MIN,YUE,WU) identification with the system of GMM show that DTFC feature is more effective than SDC feature.The recognition rate of the system with DTFC features reaches 98.89%,more than that with SDC feature by 3.1%.更多还原