基于DAE-RBM-PLDA的说话人确认信道补偿技术

Technology of speaker verification channel compensation based on DAE-RBM-PLDA

在说话人识别系统中,一种结合深度神经网路(DNN)、身份认证矢量(i-vector)和概率线性鉴别分析(PLDA)的模型被证明十分有效。为进一步提升PLDA模型信道补偿的性能,将降噪自动编码器(DAE)和受限玻尔兹曼机(RBM)以及它们的组合(DAE-RBM)分别应用到信道补偿PLDA模型端,降低说话人i-vector空间信道信息的影响。实验表明相比标准PLDA系统,基于DAE-PLDA和RBM-PLDA的识别系统的等错误率(EER)和检测代价函数(DCF)都显著降低,结合两者优势的DAE-RBMPLDA使系统识别性能得到了进一步提升。

A hybrid model combining the deep neural network( DNN),i-vector and probabilistic linear discriminant analysis( PLDA) has been shown effective in the system of speaker recognition. In order to improve the performance of PLDA recognition model,the denoising autoencoder( DAE) and restricted boltzmann machine( RBM) and the combination of them( DAE-RBM) are used to channel compensation on PLDA model to minimize the effect of the speaker i-vector space channel information. The experiment showed that the recognition system based on DAE-PLDA and RBM-PLDA is significantly decreased than the standard PLDA for the equal error rate( EER) and detection function( DCF).The DAE-RBM-PLDA which combined with the advantages of them makes the performance of the recognition system has been further improved.