Less is More: Improved RNN-T Decoding Using Limited Label Context and Path Merging
Authors:
- 6 June 2021
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 5659-5663
Abstract
End-to-end models that condition the output sequence on all previously predicted labels have emerged as popular alternatives to conventional systems for automatic speech recognition (ASR). Since distinct label histories correspond to distinct models states, such models are decoded using an approximate beam-search which produces a tree of hypotheses.In this work, we study the influence of the amount of label context on the model’s accuracy, and its impact on the efficiency of the decoding process. We find that we can limit the context of the recurrent neural network transducer (RNN-T) during training to just four previous word-piece labels, without degrading word error rate (WER) relative to the full-context baseline. Limiting context also provides opportunities to improve decoding efficiency by removing redundant paths from the active beam, and instead retaining them in the final lattice. This path-merging scheme can also be applied when decoding the baseline full-context model through an approximation. Overall, we find that the proposed path-merging scheme is extremely effective, allowing us to improve oracle WERs by up to 36% over the baseline, while simultaneously reducing the number of model evaluations by up to 5.3% without any degradation in WER, or up to 15.7% when lattice rescoring is applied.Keywords
All Related Versions
This publication has 26 references indexed in Scilit:
- Towards Better Decoding and Language Model Integration in Sequence to Sequence Models2017
- Generation of Large-Scale Simulated Utterances in Virtual Rooms to Train Deep-Neural Networks for Far-Field Speech Recognition in Google Home2017
- In-Datacenter Performance Analysis of a Tensor Processing Unit2017
- Listen, attend and spell: A neural network for large vocabulary conversational speech recognition2016
- Efficient lattice rescoring using recurrent neural network language models2014
- Speech recognition with deep recurrent neural networks2013
- Spoken language understandingIEEE Signal Processing Magazine, 2008
- Connectionist temporal classification2006
- Long Short-Term MemoryNeural Computation, 1997
- Continuous speech recognitionIEEE Signal Processing Magazine, 1995