arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2211.09313eess.AScs.SD
收藏

基于无监督模型的端到端无格MMI说话人自适应语音识别模型

Unsupervised Model-based speaker adaptation of end-to-end lattice-free MMI model for speech recognition

Institute of Software, Chinese Academy of Sciences(中国科学院软件研究所);The Chinese University of Hong Kong(香港中文大学)

Xurong Xie, Xunying Liu, Hui Chen, Hongan Wang

摘要:

说话人变异性建模是自动语音识别(ASR)系统面临的一个关键挑战。在本文中,学习隐藏单元贡献(LHUC)为基础的自适应技术与紧凑的扬声器相关(SD)参数,以促进两个扬声器自适应训练(SAT)和无监督测试时扬声器自适应的端到端(E2 E)无格MMI(LF-MMI)模型。提出了一种基于无监督模型的自适应框架,利用LF-MMI和交叉熵准则估计E2 E范式中的SD参数。标准LHUC自适应的各种正则化方法,例如,贝叶斯LHUC(BLHUC)的适应,系统地研究,以减轻过度拟合的风险,在E2 E LF-MMI CNN-TDNN和CNN-TDNN-BLSTM模型。基于格的置信度估计用于自适应数据选择,以减少监督标签的不确定性。在300小时的交换机任务上的实验表明,在所提出的无监督E2 E自适应框架中将BLHUC应用于基于字节对编码(BPE)的E2 E LF-MMI系统在NIST Hub 5 '00和RT 03评估集上的相对字错误率(WER)降低高达10.5%和14.7%,分别在9.0%和9.7%的WER中获得最佳性能。这些结果可与现有技术的适配的LF-MMI混杂系统和适配的基于Conformer的E2 E系统的结果相比较。

英文摘要:

Modeling the speaker variability is a key challenge for automatic speech recognition (ASR) systems. In this paper, the learning hidden unit contributions (LHUC) based adaptation techniques with compact speaker dependent (SD) parameters are used to facilitate both speaker adaptive training (SAT) and unsupervised test-time speaker adaptation for end-to-end (E2E) lattice-free MMI (LF-MMI) models. An unsupervised model-based adaptation framework is proposed to estimate the SD parameters in E2E paradigm using LF-MMI and cross entropy (CE) criterions. Various regularization methods of the standard LHUC adaptation, e.g., the Bayesian LHUC (BLHUC) adaptation, are systematically investigated to mitigate the risk of overfitting, on E2E LF-MMI CNN-TDNN and CNN-TDNN-BLSTM models. Lattice-based confidence score estimation is used for adaptation data selection to reduce the supervision label uncertainty. Experiments on the 300-hour Switchboard task suggest that applying BLHUC in the proposed unsupervised E2E adaptation framework to byte pair encoding (BPE) based E2E LF-MMI systems consistently outperformed the baseline systems by relative word error rate (WER) reductions up to 10.5% and 14.7% on the NIST Hub5'00 and RT03 evaluation sets, and achieved the best performance in WERs of 9.0% and 9.7%, respectively. These results are comparable to the results of state-of-the-art adapted LF-MMI hybrid systems and adapted Conformer-based E2E systems.

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。

↑