Feature engineering in Context-Dependent Deep Neural Networks for conversational speech transcription

Author

Seide, Frank ; Li, Gang ; Chen, Xie ; Yu, Dong

Author_Institution

Microsoft Res. Asia, Beijing, China

fYear

2011

fDate

11-15 Dec. 2011

Firstpage

24

Lastpage

29

Abstract

We investigate the potential of Context-Dependent Deep-Neural-Network HMMs, or CD-DNN-HMMs, from a feature-engineering perspective. Recently, we had shown that for speaker-independent transcription of phone calls (NIST RT03S Fisher data), CD-DNN-HMMs reduced the word error rate by as much as one third-from 27.4%, obtained by discriminatively trained Gaussian-mixture HMMs with HLDA features, to 18.5%-using 300+ hours of training data (Switchboard), 9000+ tied triphone states, and up to 9 hidden network layers.

Keywords

Gaussian processes; hidden Markov models; neural nets; speaker recognition; speech synthesis; Gaussian-mixture HMM; context-dependent deep neural networks; conversational speech transcription; hidden Markov model; speaker-independent transcription; speech-to-text transcription; tied triphone states; word error rate; Accuracy; Adaptation models; Feature extraction; Hidden Markov models; Training; Transforms; Vectors;

fLanguage

English

Publisher

ieee

Conference_Titel

Automatic Speech Recognition and Understanding (ASRU), 2011 IEEE Workshop on

Conference_Location

Waikoloa, HI

Print_ISBN

978-1-4673-0365-1

Electronic_ISBN

978-1-4673-0366-8

Type

conf

DOI

10.1109/ASRU.2011.6163899

Filename

6163899