Improving End-to-End Speech Recognition with Policy Learning
This work addresses a key training inefficiency in speech recognition systems, offering incremental improvements for researchers and practitioners in the field.
The paper tackles the mismatch between the maximum likelihood objective and word error rate metric in end-to-end speech recognition by jointly training with maximum likelihood and policy gradient, resulting in relative performance improvements of 4% to 13% and achieving WERs of 5.53% on Wall Street Journal and 5.42% and 14.70% on Librispeech datasets.
Connectionist temporal classification (CTC) is widely used for maximum likelihood learning in end-to-end speech recognition models. However, there is usually a disparity between the negative maximum likelihood and the performance metric used in speech recognition, e.g., word error rate (WER). This results in a mismatch between the objective function and metric during training. We show that the above problem can be mitigated by jointly training with maximum likelihood and policy gradient. In particular, with policy learning we are able to directly optimize on the (otherwise non-differentiable) performance metric. We show that joint training improves relative performance by 4% to 13% for our end-to-end model as compared to the same model learned through maximum likelihood. The model achieves 5.53% WER on Wall Street Journal dataset, and 5.42% and 14.70% on Librispeech test-clean and test-other set, respectively.