Cross-language transfer speech recognition using deep learning

Cross-language transfer speech recognition aims to transform phoneme models for a source language to recognize a target language lacking labeled data and other linguistic resources. In this paper, sparse auto-encoder, a deep learning method, is introduced to derive shared speech features between source and target language using semi-supervised learning. It can extract the shared representation of phonemes between the source and target languages so that the target phones can be mapped to the appropriate phones of the source languages. The experimental results showed this method performs better on cross-language phones recognition than the method based on multilayer perceptron.