Low-resource Uyghur Speech Recognition Based on Transfer Learning
摘要
In speech cognition,a substantial amount of data is required for acoustic model training,so the performance of the Deep Neural Network(DNN) acoustic model trained on a scarce amount of data is limited.Aiming at this problem,this paper proposes a transfer learning method to improve the Uyghur speech recognition starting from models trained by other resource-rich speech databases,so as to get a better acoustic model for Uyghur speech recognition.Experimental result shows that the above method can significantly improve the Uyghur speech recognition performance compared with the baseline models trained only on Uyghur speech database.
引用本文（GB/T 7714）
WANG Junchao,HUANG Hao,XU Haihua,HU Ying. Low-resource Uyghur Speech Recognition Based on Transfer Learning[J]. DOAJ (DOAJ: Directory of Open Access Journals), 2018.
引文网络
本站仅收录题录与摘要供学习参考，全文版权归属出版方；如有侵权请联系我们删除。