Constrained Output Embeddings for End-to-End Code-Switching Speech Recognition with Only Monolingual Data

Yerbolat Khassanov, Haihua Xu, Van Tung Pham, Zhiping Zeng, Eng Siong Chng, Chongjia Ni, Bin Ma


The lack of code-switch training data is one of the major concerns in the development of end-to-end code-switching automatic speech recognition (ASR) models. In this work, we propose a method to train an improved end-to-end code-switching ASR using only monolingual data. Our method encourages the distributions of output token embeddings of monolingual languages to be similar, and hence, promotes the ASR model to easily code-switch between languages. Specifically, we propose to use Jensen-Shannon divergence and cosine distance based constraints. The former will enforce output embeddings of monolingual languages to possess similar distributions, while the later simply brings the centroids of two distributions to be close to each other. Experimental results demonstrate high effectiveness of the proposed method, yielding up to 4.5% absolute mixed error rate improvement on Mandarin-English code-switching ASR task.


 DOI: 10.21437/Interspeech.2019-1867

Cite as: Khassanov, Y., Xu, H., Pham, V.T., Zeng, Z., Chng, E.S., Ni, C., Ma, B. (2019) Constrained Output Embeddings for End-to-End Code-Switching Speech Recognition with Only Monolingual Data. Proc. Interspeech 2019, 2160-2164, DOI: 10.21437/Interspeech.2019-1867.


@inproceedings{Khassanov2019,
  author={Yerbolat Khassanov and Haihua Xu and Van Tung Pham and Zhiping Zeng and Eng Siong Chng and Chongjia Ni and Bin Ma},
  title={{Constrained Output Embeddings for End-to-End Code-Switching Speech Recognition with Only Monolingual Data}},
  year=2019,
  booktitle={Proc. Interspeech 2019},
  pages={2160--2164},
  doi={10.21437/Interspeech.2019-1867},
  url={http://dx.doi.org/10.21437/Interspeech.2019-1867}
}