Sequence Training in DNN/CNN for ASR

最新推荐文章于 2025-05-28 15:30:53 发布

原创最新推荐文章于 2025-05-28 15:30:53 发布 · 5.3k 阅读

2 ·

CC 4.0 BY-SA版权

语音识别同时被 2 个专栏收录

4 篇文章

订阅专栏

Kaldi

3 篇文章

订阅专栏

本文介绍了深度神经网络在语音识别领域的关键研究进展，包括IBM、Microsoft的研究成果及Kaldi工具中的重要技术实现。涵盖了序列训练、最小贝叶斯风险训练、分布式Hessian-Free优化等方法。

IBM系：

【1】B. Kingsbury, “Lattice-Based Optimization of Sequence Classification Criteria for Neural-Network Acoustic Modeling,” in Proc. ICASSP,2009.

【2】Kingsbury B, Sainath T N, Soltau H. Scalable Minimum Bayes Risk Training of Deep Neural Network Acoustic Models Using Distributed Hessian-free Optimization[C]//INTERSPEECH. 2012.

Microsoft:

【1】Mohamed A, Yu D, Deng L. Investigation of full-sequence training of deep belief networks for speech recognition[C]//INTERSPEECH. 2010: 2846-2849.

【2】Su H, Li G, Yu D, et al. Error back propagation for sequence training of context-dependent deep networks for conversational speech transcription[C]//Proc. ICASSP. 2013.

Kaldi相关：

【1】Veselý K, Ghoshal A, Burget L, et al. Sequence discriminative training of deep neural networks[C]//Proc. INTERSPEECH. 2013: 2345-2349.

其他：

【1】Tachioka Y, Watanabe S. Discriminative training of acoustic models for system combination[C]//Proc. INTERSPEECH. 2013: 2355-2359.