arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2103.14152eess.AScs.CLcs.LGcs.SD
收藏

基于剩余能量的端到端语音识别模型

Residual Energy-Based Models for End-to-End Speech Recognition

Google(谷歌)

Qiujia Li, Yu Zhang, Bo Li, Liangliang Cao, Philip C. Woodland

摘要:

具有自回归解码器的端到端模型在自动语音识别(ASR)方面显示出令人印象深刻的结果。这些模型将序列级概率公式化为给定其历史的所有单个令牌的条件概率的乘积。然而,由于暴露偏差等因素,局部归一化模型的性能可能是次优的。因此,模型分布不同于底层数据分布。在本文中,基于剩余能量的模型(R-EBM)被提出来补充自回归ASR模型,以缩小这两个分布之间的差距。同时,R-EBM也可以被看作是话语级的置信度估计器,这可能有利于许多下游任务。在100小时的LibriSpeech数据集上的实验表明,R-EBM可以降低8.2%/6.7%的单词错误率,同时在测试干净/测试其他集上提高12.6%/28.4%的精确度-召回率置信度曲线下的面积。此外,在使用自监督学习(wav 2 vec 2.0)的最先进模型上,R-EBM仍然显着提高了WER和置信度估计性能。

英文摘要:

End-to-end models with auto-regressive decoders have shown impressive results for automatic speech recognition (ASR). These models formulate the sequence-level probability as a product of the conditional probabilities of all individual tokens given their histories. However, the performance of locally normalised models can be sub-optimal because of factors such as exposure bias. Consequently, the model distribution differs from the underlying data distribution. In this paper, the residual energy-based model (R-EBM) is proposed to complement the auto-regressive ASR model to close the gap between the two distributions. Meanwhile, R-EBMs can also be regarded as utterance-level confidence estimators, which may benefit many downstream tasks. Experiments on a 100hr LibriSpeech dataset show that R-EBMs can reduce the word error rates (WERs) by 8.2%/6.7% while improving areas under precision-recall curves of confidence scores by 12.6%/28.4% on test-clean/test-other sets. Furthermore, on a state-of-the-art model using self-supervised learning (wav2vec 2.0), R-EBMs still significantly improves both the WER and confidence estimation performance.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑