Improving Streaming Transformer Based ASR Under a Framework of Self-supervised Learning

被引:3
作者
Cao, Songjun [1 ]
Kang, Yueteng [1 ]
Fu, Yanzhe [1 ]
Xu, Xiaoshuo [1 ]
Sun, Sining [1 ]
Zhang, Yike [1 ]
Ma, Long [1 ]
机构
[1] Tencent Cloud Xiaowei, Beijing, Peoples R China
来源
INTERSPEECH 2021 | 2021年
关键词
speech recognition; streaming transformer; self-supervised learning; knowledge distilling;
D O I
10.21437/Interspeech.2021-1454
中图分类号
R36 [病理学]; R76 [耳鼻咽喉科学];
学科分类号
100104 ; 100213 ;
摘要
Recently self-supervised learning has emerged as an effective approach to improve the performance of automatic speech recognition (ASR). Under such a framework, the neural network is usually pre-trained with massive unlabeled data and then fine-tuned with limited labeled data. However, the non-streaming architecture like bidirectional transformer is usually adopted by the neural network to achieve competitive results, which can not be used in streaming scenarios. In this paper, we mainly focus on improving the performance of streaming transformer under the self-supervised learning framework. Specifically, we propose a novel two-stage training method during fine-tuning, which combines knowledge distilling and self-training. The proposed training method achieves 16.3% relative word error rate (WER) reduction on Librispeech noisy test set. Finally, by only using the 100h clean subset of Librispeech as the labeled data and the rest (860h) as the unlabeled data, our streaming transformer based model obtains competitive WERs 3.5/8.7 on Librispeech clean/noisy test sets.
引用
收藏
页码:706 / 710
页数:5
相关论文
共 29 条
  • [1] Baevski A., 2020, ADV NEURAL INFORM PR
  • [2] Baevski A, 2020, INT CONF ACOUST SPEE, P7694, DOI [10.1109/icassp40776.2020.9054224, 10.1109/ICASSP40776.2020.9054224]
  • [3] Baevski Alexei, 2020, P ICLR
  • [4] Chen T, 2020, PR MACH LEARN RES, V119
  • [5] Chen X, 2021, Arxiv, DOI [arXiv:2010.11395, 10.48550/arXiv.2010.11395]
  • [6] Devlin J, 2019, Arxiv, DOI [arXiv:1810.04805, DOI 10.48550/ARXIV.1810.04805]
  • [7] Dong LH, 2019, INT CONF ACOUST SPEE, P5656, DOI [10.1109/ICASSP.2019.8682954, 10.1109/icassp.2019.8682954]
  • [8] Peters ME, 2018, Arxiv, DOI [arXiv:1802.05365, 10.48550/ARXIV.1802.05365]
  • [9] Graves A., 2006, INT C MACH LEARN
  • [10] Henaff OJ, 2020, PR MACH LEARN RES, V119