Investigation on Estimation of Sentence Probability by Combining Forward, Backward and Bi-directional LSTM-RNNs

Kazuki Irie, Zhihong Lei, Liuhui Deng, Ralf Schlüter, Hermann Ney


A combination of forward and backward long short-term memory (LSTM) recurrent neural network (RNN) language models is a popular model combination approach to improve the estimation of the sequence probability in the second pass N-best list rescoring in automatic speech recognition (ASR). In this work, we further push such an idea by proposing a combination of three models: a forward LSTM language model, a backward LSTM language model and a bi-directional LSTM based gap completion model. We derive such a combination method from a forward backward decomposition of the sequence probability. We carry out experiments on the Switchboard speech recognition task. While we empirically find that such a combination gives slight improvements in perplexity over the combination of forward and backward models, we finally show that a combination of the same number of forward models gives the best perplexity and word error rate (WER) overall.


 DOI: 10.21437/Interspeech.2018-1766

Cite as: Irie, K., Lei, Z., Deng, L., Schlüter, R., Ney, H. (2018) Investigation on Estimation of Sentence Probability by Combining Forward, Backward and Bi-directional LSTM-RNNs. Proc. Interspeech 2018, 392-395, DOI: 10.21437/Interspeech.2018-1766.


@inproceedings{Irie2018,
  author={Kazuki Irie and Zhihong Lei and Liuhui Deng and Ralf Schlüter and Hermann Ney},
  title={Investigation on Estimation of Sentence Probability by Combining Forward, Backward and Bi-directional LSTM-RNNs},
  year=2018,
  booktitle={Proc. Interspeech 2018},
  pages={392--395},
  doi={10.21437/Interspeech.2018-1766},
  url={http://dx.doi.org/10.21437/Interspeech.2018-1766}
}