Lattice Rescoring Strategies for Long Short Term Memory Language Models in Speech Recognition
This work addresses efficiency challenges for speech recognition practitioners, but it is incremental as it builds on existing lattice-rescoring methods.
The paper tackles the problem of integrating computationally expensive LSTM language models into speech recognition systems by evaluating lattice-rescoring algorithms, resulting in an 8% relative reduction in word error rate on a YouTube task compared to N-gram models.
Recurrent neural network (RNN) language models (LMs) and Long Short Term Memory (LSTM) LMs, a variant of RNN LMs, have been shown to outperform traditional N-gram LMs on speech recognition tasks. However, these models are computationally more expensive than N-gram LMs for decoding, and thus, challenging to integrate into speech recognizers. Recent research has proposed the use of lattice-rescoring algorithms using RNNLMs and LSTMLMs as an efficient strategy to integrate these models into a speech recognition system. In this paper, we evaluate existing lattice rescoring algorithms along with new variants on a YouTube speech recognition task. Lattice rescoring using LSTMLMs reduces the word error rate (WER) for this task by 8\% relative to the WER obtained using an N-gram LM.