One solution to automatic speech recognition (ASR) of overlapping speakers is
to separate speech and then perform ASR on the separated signals. Commonly, the
separator produces artefacts which often degrade ASR performance. Addressing
this issue typically requires reference transcriptions to jointly train the
separation and ASR networks. This is often not viable for training on
real-world in-domain audio where reference transcript information is not always
available. This paper proposes a transcription-free method for joint training
using only audio signals. The proposed method uses embedding differences of
pre-trained ASR encoders as a loss with a proposed modification to permutation
invariant training (PIT) called guided PIT (GPIT). The method achieves a 6.4%
improvement in word error rate (WER) measures over a signal-level loss and also
shows enhancement improvements in perceptual measures such as short-time
objective intelligibility (STOI).

该论文提出了一种无需参考文本信息的联合训练方法，利用预训练的自动语音识别编码器的嵌入差异作为损失，通过改进的排列不变训练（PIT）方法 —— 引导式 PIT（GPIT），在不同度量指标上取得了 6.4% 的词错误率（WER）改善和感知度量指标（如短时客观清晰度）的提升。

无需转写的噪声和混响多说话人自动语音识别的语音分离模型微调

Transcription-Free Fine-Tuning of Speech Separation Models for Noisy and  Reverberant Multi-Speaker Automatic Speech Recognition

Given two languages, a separator is a third language that contains the first
one and is disjoint from the second one. We investigate the following decision
problem: given two regular input languages of finite words, decide whether
there exists a first-order definable separator. We prove that in order to
answer this question, sufficient information can be extracted from semigroups
recognizing the input languages, using a fixpoint computation. This yields an
EXPTIME algorithm for checking first-order separability. Moreover, the
correctness proof of this algorithm yields a stronger result, namely a
description of a possible separator. Finally, we generalize this technique to
answer the same question for regular languages of infinite words.

研究一种决策问题：给定两个有限单词的正则输入语言，判断是否存在第一阶可定义的分隔符；证明可以从识别输入语言的半群中提取充分信息并使用不动点计算以回答此问题，这产生了一个检查一阶可分离性的 EXPTIME 算法；进一步推广此技术以回答针对无限字的正则语言同样的问题，并得到一个可能的分隔符描述。