Because of predicting all the target tokens in parallel, the non-autoregressive models greatly improve the decoding efficiency of speech recognition compared with traditional autoregressive models. In this work, we present dynamic alignment Mask CTC, introducing two methods: (1) Aligned Cross Entropy (AXE), finding the monotonic alignment that minimizes the cross-entropy loss through dynamic programming, (2) Dynamic Rectification, creating new training samples by replacing some masks with model predicted tokens. The AXE ignores the absolute position alignment between prediction and ground truth sentence and focuses on tokens matching in relative order. The dynamic rectification method makes the model capable of simulating the non-mask but possible wrong tokens, even if they have high confidence. Our experiments on WSJ dataset demonstrated that not only AXE loss but also the rectification method could improve the WER performance of Mask CTC.

翻译：由于非自回归模型能够并行预测所有目标标记，与传统自回归模型相比，其显著提升了语音识别的解码效率。本文提出动态对齐掩码CTC（Dynamic Alignment Mask CTC），引入两种方法：（1）对齐交叉熵（Aligned Cross Entropy, AXE），通过动态规划寻找使交叉熵损失最小的单调对齐路径；（2）动态矫正（Dynamic Rectification），通过用模型预测的标记替换部分掩码来生成新的训练样本。AXE方法忽略预测序列与真实序列之间的绝对位置对齐，而专注于标记的相对顺序匹配。动态矫正方法使模型能够模拟未被掩码但可能错误的标记，即使这些标记具有高置信度。我们在WSJ数据集上的实验表明，AXE损失和矫正方法均能提升Mask CTC的词错误率（WER）性能。