Mimura et al - REVERB Workshop 2014 REVERBERANT SPEECH...

Info icon This preview shows pages 1–2. Sign up to view the full content.

REVERBERANT SPEECH RECOGNITION COMBINING DEEP NEURAL NETWORKS AND DEEP AUTOENCODERS Masato Mimura, Shinsuke Sakai, and Tatsuya Kawahara Academic Center for Computing and Media Studies, Kyoto University Sakyo-ku, Kyoto 606-8501, Japan { mimura | sakai | kawahara } @ar.media.kyoto-u.ac.jp ABSTRACT We propose an approach to reverberant speech recognition adopting deep learning in front end as well as back end of the sys- tem. At the front end, we adopt a deep autoencoder for enhancing the speech feature parameters, and the recognition is performed us- ing a DNN-HMM acoustic models trained on multi-condition data. The system was evaluated through the ASR task in Chime Challenge 2014. The DNN-HMM system trained on the multi-condition train- ing set achieved a conspicuously higher word accuracy compared to the MLLR-adapted GMM-HMM system trained on the same data. Furthermore, feature enhancement with the deep autoencoder con- tributed to the improvement of recognition accuracy especially in the more adverse conditions. When the DNN-HMM was used with- out the deep autoencoder front end, it resulted in a better perfor- mance than the non-adapted GMM-HMM system, but was not as good as the adapted GMM-HMM system. However, it outperformed the adapted GMM-HMM system when combined with the deep au- toencoder. Index Terms reverberant speech recognition, Deep Neural Net- work(DNN), deep autoencoder 1. INTRODUCTION In recent years, the speech recognition technology based on statisti- cal techniques achieved a remarkable progress supported by the ever increasing training data and the improvements in the computing re- sources. Applications such as voice search are now being used in our daily life. However, speech recognition in adverse conditions is still a difficult task and the recognition accuracies in adverse envi- ronments such as those with reverberation and background noise are still staying at low levels. A key breakthrough for speech recognition technology to be ac- cepted widely in the society will be the establishment of the method- ology for easier speech interface with hands-free input. Speech re- verberation adversely influences the speech recognition accuracy in such conditions and various efforts have been made to improve the recognition performance for the reverberant speech. Reverberant speech recognition has so far been tackled by apply- ing feature enhancement at the front end, and by attempting model adaptation and the use of more sophisticated recognition techniques. Speech enhancement techniques include deconvolution approaches that tries to reconstruct clean speech by inverse filtering the reverber- ant speech [1][2][3] and spectral enhancement approaches that esti- mate and remove the influences of the late reflection [4][5]. Since improvement measured by SNR may not be directly related to the speech recognition accuracy, there also are approaches to enhance speech based on speech recognition likelihoods in the back end [6].
Image of page 1

Info icon This preview has intentionally blurred sections. Sign up to view the full version.

Image of page 2
This is the end of the preview. Sign up to access the rest of the document.
  • Winter '14
  • Ms.Hariison
  • Speech recognition, neural network, DNNs, Word error rate, reverberant speech

{[ snackBarMessage ]}

What students are saying

  • Left Quote Icon

    As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

    Student Picture

    Kiran Temple University Fox School of Business ‘17, Course Hero Intern

  • Left Quote Icon

    I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

    Student Picture

    Dana University of Pennsylvania ‘17, Course Hero Intern

  • Left Quote Icon

    The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

    Student Picture

    Jill Tulane University ‘16, Course Hero Intern