arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2302.08579eess.AScs.SD
收藏

使用可更换内部LM和残差Softmax的自适应端到端ASR模型

Adaptable End-to-End ASR Models using Replaceable Internal LMs and Residual Softmax

University of Cambridge(剑桥大学)

Keqi Deng, Philip C. Woodland

摘要:

端到端(E2 E)自动语音识别(ASR)隐式地学习成对的音频转录训练数据的标记序列分布。然而,它仍然受到从训练到测试的领域转移的影响,领域适应仍然具有挑战性。为了解决这一问题,本文设计了一种可替换内部语言模型(RILM)方法,该方法可以在解码阶段直接用目标域LM替换E2 E ASR模型的内部语言模型(LM)。此外,本文还提出了一种残差softmax(R-softmax),该算法专为基于CTC的E2 E ASR模型设计,以适应目标域,而无需在推理过程中重新训练。对于在LibriSpeech语料库上训练的E2 E ASR模型,实验表明,所提出的方法在Switchboard数据上提供了2.6%的绝对WER减少,在AESRC 2020语料库上提供了1.0%的WER减少,同时保持了域内ASR结果。

英文摘要:

End-to-end (E2E) automatic speech recognition (ASR) implicitly learns the token sequence distribution of paired audio-transcript training data. However, it still suffers from domain shifts from training to testing, and domain adaptation is still challenging. To alleviate this problem, this paper designs a replaceable internal language model (RILM) method, which makes it feasible to directly replace the internal language model (LM) of E2E ASR models with a target-domain LM in the decoding stage when a domain shift is encountered. Furthermore, this paper proposes a residual softmax (R-softmax) that is designed for CTC-based E2E ASR models to adapt to the target domain without re-training during inference. For E2E ASR models trained on the LibriSpeech corpus, experiments showed that the proposed methods gave a 2.6% absolute WER reduction on the Switchboard data and a 1.0% WER reduction on the AESRC2020 corpus while maintaining intra-domain ASR results.

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。

↑