摘要翻译：
尽管序列到序列方法在自动语音识别(ASR)系统中取得了成功，但这些模型仍然存在一些问题，主要是由于训练条件和推理条件之间的不匹配。在序列到序列的结构中，该模型被训练以预测当前时间步长的字形，给定语音信号的输入和前一个时间步长的地面真值字形历史。然而，目前还不清楚该模型在推理过程中如何接近真实世界的语音。因此，根据先前的预测从头开始生成整个转录是复杂的，而且错误会随着时间的推移而传播。此外，该模型被优化以最大化训练数据的可能性，而不是实际量化识别质量的错误率评估指标。本文提出了一种基于强化学习(RL)的序列间ASR模型训练策略。与最大似然估计的标准训练方案不同，我们提出的方法使用策略梯度算法。我们可以（1）在训练过程中根据模型的预测对整个转录进行采样；（2）以负Levenshtein距离为奖励直接优化模型。实验结果表明，与仅用最大似然估计训练的模型相比，我们显著地提高了模型的性能。
---
英文标题：
《Sequence-to-Sequence ASR Optimization via Reinforcement Learning》
---
作者：
Andros Tjandra, Sakriani Sakti, Satoshi Nakamura
---
最新提交年份：
2018
---
分类信息：

一级分类：Computer Science        计算机科学
二级分类：Computation and Language        计算与语言
分类描述：Covers natural language processing. Roughly includes material in ACM Subject Class I.2.7. Note that work on artificial languages (programming languages, logics, formal systems) that does not explicitly address natural-language issues broadly construed (natural-language processing, computational linguistics, speech, text retrieval, etc.) is not appropriate for this area.
涵盖自然语言处理。大致包括ACM科目I.2.7类的材料。请注意，人工语言（编程语言、逻辑学、形式系统）的工作，如果没有明确地解决广义的自然语言问题（自然语言处理、计算语言学、语音、文本检索等），就不适合这个领域。
--
一级分类：Computer Science        计算机科学
二级分类：Machine Learning        机器学习
分类描述：Papers on all aspects of machine learning research (supervised, unsupervised, reinforcement learning, bandit problems, and so on) including also robustness, explanation, fairness, and methodology. cs.LG is also an appropriate primary category for applications of machine learning methods.
关于机器学习研究的所有方面的论文（有监督的，无监督的，强化学习，强盗问题，等等），包括健壮性，解释性，公平性和方法论。对于机器学习方法的应用，CS.LG也是一个合适的主要类别。
--
一级分类：Computer Science        计算机科学
二级分类：Sound        声音
分类描述：Covers all aspects of computing with sound, and sound as an information channel. Includes models of sound, analysis and synthesis, audio user interfaces, sonification of data, computer music, and sound signal processing. Includes ACM Subject Class H.5.5, and intersects with H.1.2, H.5.1, H.5.2, I.2.7, I.5.4, I.6.3, J.5, K.4.2.
涵盖了声音计算的各个方面，以及声音作为一种信息通道。包括声音模型、分析和合成、音频用户界面、数据的可听化、计算机音乐和声音信号处理。包括ACM学科类H.5.5，并与H.1.2、H.5.1、H.5.2、I.2.7、I.5.4、I.6.3、J.5、K.4.2交叉。
--
一级分类：Electrical Engineering and Systems Science        电气工程与系统科学
二级分类：Audio and Speech Processing        音频和语音处理
分类描述：Theory and methods for processing signals representing audio, speech, and language, and their applications. This includes analysis, synthesis, enhancement, transformation, classification and interpretation of such signals as well as the design, development, and evaluation of associated signal processing systems. Machine learning and pattern analysis applied to any of the above areas is also welcome.  Specific topics of interest include: auditory modeling and hearing aids; acoustic beamforming and source localization; classification of acoustic scenes; speaker separation; active noise control and echo cancellation; enhancement; de-reverberation; bioacoustics; music signals analysis, synthesis and modification; music information retrieval;  audio for multimedia and joint audio-video processing; spoken and written language modeling, segmentation, tagging, parsing, understanding, and translation; text mining; speech production, perception, and psychoacoustics; speech analysis, synthesis, and perceptual modeling and coding; robust speech recognition; speaker recognition and characterization; deep learning, online learning, and graphical models applied to speech, audio, and language signals; and implementation aspects ranging from system architecture to fast algorithms.
处理代表音频、语音和语言的信号的理论和方法及其应用。这包括分析、合成、增强、转换、分类和解释这些信号，以及相关信号处理系统的设计、开发和评估。机器学习和模式分析应用于上述任何领域也是受欢迎的。感兴趣的具体主题包括：听觉建模和助听器；声波束形成与声源定位；声场景分类；说话人分离；有源噪声控制和回声消除；增强；去混响；生物声学；音乐信号的分析、合成与修饰；音乐信息检索；多媒体音频和联合音视频处理；口语和书面语建模、切分、标注、句法分析、理解和翻译；文本挖掘；言语产生、感知和心理声学；语音分析、合成、感知建模和编码；鲁棒语音识别；说话人识别与特征描述；应用于语音、音频和语言信号的深度学习、在线学习和图形模型；以及从系统架构到快速算法的实现方面。
--

---
英文摘要：
  Despite the success of sequence-to-sequence approaches in automatic speech recognition (ASR) systems, the models still suffer from several problems, mainly due to the mismatch between the training and inference conditions. In the sequence-to-sequence architecture, the model is trained to predict the grapheme of the current time-step given the input of speech signal and the ground-truth grapheme history of the previous time-steps. However, it remains unclear how well the model approximates real-world speech during inference. Thus, generating the whole transcription from scratch based on previous predictions is complicated and errors can propagate over time. Furthermore, the model is optimized to maximize the likelihood of training data instead of error rate evaluation metrics that actually quantify recognition quality. This paper presents an alternative strategy for training sequence-to-sequence ASR models by adopting the idea of reinforcement learning (RL). Unlike the standard training scheme with maximum likelihood estimation, our proposed approach utilizes the policy gradient algorithm. We can (1) sample the whole transcription based on the model's prediction in the training process and (2) directly optimize the model with negative Levenshtein distance as the reward. Experimental results demonstrate that we significantly improved the performance compared to a model trained only with maximum likelihood estimation.
---
PDF链接：
https://arxiv.org/pdf/1710.10774