Unsupervised training scheme with non-stereo data for empirical feature vector compensation


In this paper, a novel training scheme based on unsupervised and non-stereo data is presented for Multi-Environment Modelbased LInear Normalization (MEMLIN) and MEMLIN with cross-probability model based on GMMs (MEMLIN-CPM). Both are data-driven feature vector normalization techniques which have been proved very effective in dynamic noisy acoustic… (More)


2 Figures and Tables

Slides referencing similar topics