A Semisupervised Approach for Language Identification based on Ladder Networks

摘要

In this study we address the problem of training a neuralnetwork for languageidentification using both labeled and unlabeled speech samples in the form ofi-vectors. We propose a neural network architecture that can also handleout-of-set languages. We utilize a modified version of the recently proposedLadder Network semisupervised training procedure that optimizes thereconstruction costs of a stack of denoising autoencoders. We show that thisapproach can be successfully applied to the case where the training dataset iscomposed of both labeled and unlabeled acoustic data. The results show enhancedlanguage identification on the NIST 2015 language identification dataset.