The 5th CHiME Speech Separation and Recognition Challenge (CHiME-5) [1] considers the problem of distant multi-microphone conversational speech recognition in everyday home environments. In this challenge, we take advantage of several beamforming techniques, powerful TDNN-F [2] acoustic models, pruned lattice-rescoring algorithm as well as ROVER [3] based system fusion. Compared to the official base-line, our best system achieves around 18% and 17% absolute WER reduction on the development and test sets respectively.
Cite as: Zhao, Z., Wu, J., Xie, L. (2018) The NWPU System for CHiME-5 Challenge. Proc. 5th International Workshop on Speech Processing in Everyday Environments (CHiME 2018), 16-18, doi: 10.21437/CHiME.2018-4
@inproceedings{zhao18_chime, author={Zhiwei Zhao and Jian Wu and Lei Xie}, title={{The NWPU System for CHiME-5 Challenge}}, year=2018, booktitle={Proc. 5th International Workshop on Speech Processing in Everyday Environments (CHiME 2018)}, pages={16--18}, doi={10.21437/CHiME.2018-4} }