arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 1905.03072cs.CLcs.SDeess.AS
收藏

RWTH ASR系统用于LibriSpeech:混合与注意力--无数据增强

RWTH ASR Systems for LibriSpeech: Hybrid vs Attention -- w/o Data Augmentation

RWTH(亚琛工业大学)

Christoph Lüscher, Eugen Beck, Kazuki Irie, Markus Kitza, Wilfried Michel, Albert Zeyer, Ralf Schlüter, Hermann Ney

摘要:

我们提出了最先进的自动语音识别(ASR)系统采用标准的混合DNN/HMM架构相比,基于注意力的编码器-解码器设计的LibriSpeech任务。系统开发的详细描述,包括模型设计,预训练方案,培训计划,和优化方法提供了两个系统架构。混合DNN/HMM和基于注意力的系统都采用双向LSTM进行声学建模/编码。对于语言建模,我们采用基于LSTM和Transformer的架构。我们所有的系统都使用RWTH的开源工具包RASR和RETURNN构建。据作者所知,在完整的LibriSpeech训练集上训练时获得的结果是目前发表的最好的,无论是对于混合DNN/HMM还是基于注意力的系统。我们的单混合动力系统甚至优于以前的结果,从8个单一的系统相结合。我们的比较表明,在LibriSpeech 960 h任务上,混合DNN/HMM系统在单词出错率方面比基于注意力的系统在干净的测试集上表现出15%的相对优势,在其他测试集上表现出40%的相对优势。此外,在LibriSpeech训练语料库的减少的100 h子集上的实验甚至显示出混合DNN/HMM和基于注意力的架构之间的更明显的裕度。

英文摘要:

We present state-of-the-art automatic speech recognition (ASR) systems employing a standard hybrid DNN/HMM architecture compared to an attention-based encoder-decoder design for the LibriSpeech task. Detailed descriptions of the system development, including model design, pretraining schemes, training schedules, and optimization approaches are provided for both system architectures. Both hybrid DNN/HMM and attention-based systems employ bi-directional LSTMs for acoustic modeling/encoding. For language modeling, we employ both LSTM and Transformer based architectures. All our systems are built using RWTHs open-source toolkits RASR and RETURNN. To the best knowledge of the authors, the results obtained when training on the full LibriSpeech training set, are the best published currently, both for the hybrid DNN/HMM and the attention-based systems. Our single hybrid system even outperforms previous results obtained from combining eight single systems. Our comparison shows that on the LibriSpeech 960h task, the hybrid DNN/HMM system outperforms the attention-based system by 15% relative on the clean and 40% relative on the other test sets in terms of word error rate. Moreover, experiments on a reduced 100h-subset of the LibriSpeech training corpus even show a more pronounced margin between the hybrid DNN/HMM and attention-based architectures.

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。

↑