arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2407.04439eess.AS
收藏

XLSR-Transducer:面向自监督预训练模型的流式自动语音识别

XLSR-Transducer: Streaming ASR for Self-Supervised Pretrained Models

Shashi Kumar, Srikanth Madikeri, Juan Zuluaga-Gomez, Esaú Villatoro-Tello, Iuliia Thorbecke, Petr Motlicek, Manjunath K E, Aravind Ganapathiraju

AI总结:

本文提出 XLSR-Transducer,将 XLSR-53 作为 transducer 编码器,并通过注意力掩码与 attention sinks 支持流式识别,在 AMI 和 CommonVoice 低资源场景中显著降低 WER。

摘要:

自监督预训练模型在微调后于自动语音识别任务中表现出具有竞争力的性能,即使领域内监督数据有限也是如此。然而,流行的预训练模型并不适合流式 ASR,因为它们是使用全注意力上下文训练的。本文提出 XLSR-Transducer,其中 XLSR-53 模型在 transducer 设置中用作编码器。我们在 AMI 数据集上的实验表明,与 Whisper large-v2 相比,XLSR-Transducer 的 WER 绝对提升 4%,与从零开始训练的 Zipformer transducer 模型相比则提升 8%。为实现流式能力,我们研究了 XLSR-53 模型中 transformer 层自注意力计算的不同注意力掩码模式。我们在低资源场景下于 AMI 和 CommonVoice 的 5 种语言上验证了 XLSR-Transducer。最后,通过引入 attention sinks,我们将左侧上下文减少一半,同时实现 12% 的相对 WER 改善。

英文摘要:

Self-supervised pretrained models exhibit competitive performance in automatic speech recognition on finetuning, even with limited in-domain supervised data. However, popular pretrained models are not suitable for streaming ASR because they are trained with full attention context. In this paper, we introduce XLSR-Transducer, where the XLSR-53 model is used as encoder in transducer setup. Our experiments on the AMI dataset reveal that the XLSR-Transducer achieves 4% absolute WER improvement over Whisper large-v2 and 8% over a Zipformer transducer model trained from scratch. To enable streaming capabilities, we investigate different attention masking patterns in the self-attention computation of transformer layers within the XLSR-53 model. We validate XLSR-Transducer on AMI and 5 languages from CommonVoice under low-resource scenarios. Finally, with the introduction of attention sinks, we reduce the left context by half while achieving a relative 12% improvement in WER.

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。

↑