Character-based Neural Network Language Models (NNLM) have the advantage of smaller vocabulary and faster training times in comparison to NNLMs based on multi-character units . However, applying cross-lingual transfer to character NNLM is not as straightforward . We show that only pretraining with a related language improves the ASR performance, and using an unrelated language may deteriorate it . We evaluate this aspect on ASR tasks for two target languages: Finnish (with English and Estonian as source) and Swedish (with Danish, Norwegian, and English as source), we also observe that the benefits are larger when there is much lesser target data than source data . We also show that the . benefits are . larger when . there is a much lesser . target data . than the source data than the . source data, or there is less target data, source data. We also observe the benefits of the benefits were larger than the target data.

Links: PDF - Abstract

Code :


Keywords : data - source - language - target - benefits -

Leave a Reply

Your email address will not be published. Required fields are marked *


Enjoy this blog? Please spread the word :)