# Bridging the gap between streaming and non-streaming ASR systems bydistilling ensembles of CTC and RNN-T models

arXiv: https://arxiv.org/abs/2104.14346 · HTML: https://mlsanity.com/paper/2104.14346/

## Key facts

- Cited by 3 papers in the map (data through 14 Oct 2026).
- Builds on 18 of its 29 references that are in the map.
- First posted on arXiv 25 Apr 2021.

## Builds on (most cited first, 18 of 18)

- [Conformer: Convolution-augmented Transformer for Speech Recognition](https://mlsanity.com/paper/2005.08100/) (2020)
- [Listen, Attend and Spell](https://mlsanity.com/paper/1508.01211/) (2015)
- [Attention-Based Models for Speech Recognition](https://mlsanity.com/paper/1506.07503/) (2015)
- [Transformer Transducer: A Streamable Speech Recognition Model with Transformer Encoders and RNN-T Loss](https://mlsanity.com/paper/2002.02562/) (2020)
- [Streaming End-to-end Speech Recognition For Mobile Devices](https://mlsanity.com/paper/1811.06621/) (2018)
- [Pushing the Limits of Semi-Supervised Learning for Automatic Speech Recognition](https://mlsanity.com/paper/2010.10504/) (2020)
- [Improved Noisy Student Training for Automatic Speech Recognition](https://mlsanity.com/paper/2005.09629/) (2020)
- [A Streaming On-Device End-to-End Model Surpassing Server-Side Conventional Model Quality and Latency](https://mlsanity.com/paper/2003.12710/) (2020)
- [ContextNet: Improving Convolutional Neural Networks for Automatic Speech Recognition with Global Context](https://mlsanity.com/paper/2005.03191/) (2020)
- [Recognizing long-form speech using streaming end-to-end models](https://mlsanity.com/paper/1910.11455/) (2019)
- [Streaming automatic speech recognition with the transformer model](https://mlsanity.com/paper/2001.02674/) (2020)
- [Transformer-Transducer: End-to-End Speech Recognition with Self-Attention](https://mlsanity.com/paper/1910.12977/) (2019)
- [Improving RNN Transducer Modeling for End-to-End Speech Recognition](https://mlsanity.com/paper/1909.12415/) (2019)
- [Two-Pass End-to-End Speech Recognition](https://mlsanity.com/paper/1908.10992/) (2019)
- [Towards Fast and Accurate Streaming End-to-End ASR](https://mlsanity.com/paper/2004.11544/) (2020)
- [Transformer ASR with Contextual Block Processing](https://mlsanity.com/paper/1910.07204/) (2019)
- [RNN-T Models Fail to Generalize to Out-of-Domain Audio: Causes and Solutions](https://mlsanity.com/paper/2005.03271/) (2020)
- [Improved training for online end-to-end speech recognition systems](https://mlsanity.com/paper/1711.02212/) (2017)

## Built on by (most cited first, 3 of 3)

- [BigSSL: Exploring the Frontier of Large-Scale Semi-Supervised Learning for Automatic Speech Recognition](https://mlsanity.com/paper/2109.13226/) (2021)
- [Input Length Matters: Improving RNN-T and MWER Training for Long-form Telephony Speech Recognition](https://mlsanity.com/paper/2110.03841/) (2021)
- [Exploring Attention Map Reuse for Efficient Transformer Neural Networks](https://mlsanity.com/paper/2301.12444/) (2023)

## Abstract (arXiv)

Streaming end-to-end automatic speech recognition (ASR) systems are widely used in everyday applications that require transcribing speech to text in real-time. Their minimal latency makes them suitable for such tasks. Unlike their non-streaming counterparts, streaming models are constrained to be causal with no future context and suffer from higher word error rates (WER). To improve streaming models, a recent study [1] proposed to distill a non-streaming teacher model on unsupervised utterances, and then train a streaming student using the teachers' predictions. However, the performance gap between teacher and student WERs remains high. In this paper, we aim to close this gap by using a diversified set of non-streaming teacher models and combining them using Recognizer Output Voting Error Reduction (ROVER). In particular, we show that, despite being weaker than RNN-T models, CTC models are remarkable teachers. Further, by fusing RNN-T and CTC models together, we build the strongest teachers. The resulting student models drastically improve upon streaming models of previous work [1]: the WER decreases by 41% on Spanish, 27% on Portuguese, and 13% on French.