mlsanityData to 14 Oct 2026

Edit Distance based RL for RNNT decoding

arXiv 2306.01789 ↗First posted 31 May 2023cs.SD cs.CL eess.AScited by 0

Not on the map. Why?

Abstract (arXiv)

RNN-T is currently considered the industry standard in ASR due to its exceptional WERs in various benchmark tests and its ability to support seamless streaming and longform transcription. However, its biggest drawback lies in the significant discrepancy between its training and inference objectives. During training, RNN-T maximizes all alignment probabilities by teacher forcing, while during inference, it uses beam search which may not necessarily find the maximum probable alignment. Additionally, RNN-T's inability to experience mistakes during teacher forcing training makes it more problematic when a mistake occurs in inference. To address this issue, this paper proposes a Reinforcement Learning method that minimizes the gap between training and inference time. Our Edit Distance based RL (EDRL) approach computes rewards based on the edit distance, and trains the network at every action level. The proposed approach yielded SoTA WERs on LibriSpeech for the 600M Conformer RNN-T model.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 26 of its 35 references that are in the map. First posted on arXiv 31 May 2023.

Builds on

26 of 35 references.

References · 26

  1. 2022Training language models to follow instructions with human feedbackcited by 15,971
  2. 2022Robust Speech Recognition via Large-Scale Weak Supervisioncited by 4,057
  3. 2015High-Dimensional Continuous Control Using Generalized Advantage Estimationcited by 2,026
  4. 2020Conformer: Convolution-augmented Transformer for Speech Recognitioncited by 1,863
  5. 2015Deep Reinforcement Learning with Double Q-learningcited by 1,745
  6. 2016Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translationcited by 1,475
  7. 2015Scheduled Sampling for Sequence Prediction with Recurrent Neural Networkscited by 803
  8. 2017In-Datacenter Performance Analysis of a Tensor Processing Unitcited by 618
  9. 2015Listen, Attend and Spellcited by 543
  10. 2020MLS: A Large-Scale Multilingual Dataset for Speech Researchcited by 476
  11. 2019Libri-Light: A Benchmark for ASR with Limited or No Supervisioncited by 467
  12. 2021W2v-BERT: Combining Contrastive Learning and Masked Language Modeling for Self-Supervised Speech Pre-Trainingcited by 347
14 more
  1. 2017State-of-the-art Speech Recognition With Sequence-to-Sequence Modelscited by 196
  2. 2020Pushing the Limits of Semi-Supervised Learning for Automatic Speech Recognitioncited by 187
  3. 2020Improved Noisy Student Training for Automatic Speech Recognitioncited by 141
  4. 2020A Streaming On-Device End-to-End Model Surpassing Server-Side Conventional Model Quality and Latencycited by 100
  5. 2020Self-training and Pre-training are Complementary for Speech Recognitioncited by 99
  6. 2021BigSSL: Exploring the Frontier of Large-Scale Semi-Supervised Learning for Automatic Speech Recognitioncited by 88
  7. 2019Lingvo: a Modular and Scalable Framework for Sequence-to-Sequence Modelingcited by 86
  8. 2021SpeechStew: Simply Mix All Available Speech Recognition Data to Train One Large Neural Networkcited by 83
  9. 2019Recognizing long-form speech using streaming end-to-end modelscited by 77
  10. 2020A Better and Faster End-to-End Model for Streaming ASRcited by 69
  11. 2017Minimum Word Error Rate Training for Attention-based Sequence-to-Sequence Modelscited by 68
  12. 2018Variance Reduction for Policy Gradient with Action-Dependent Factorized Baselinescited by 68
  13. 2022Pseudo Label Is Better Than Human Labelcited by 25
  14. 2018Optimal Completion Distillation for Sequence Learningcited by 14

Built on by

Not cited yet.