Reverberation affects perceived quality and intelligibility of speech signals, as well as the performance of automatic speech recognition systems. Having access to room acoustics characteristics of an environment may be used to improve speech processing systems, but such information is rarely available and in most cases has to be estimated blindly. Techniques based on the effects of reverberation on the modulation spectrum have been explored in the past, but they rely on its long-term average and do not use any information related to its temporal dynamics. In this paper, we aim to extract this information from the modulation spectrum time-series by using a deep recurrent neural network. We show the proposed model outperforms state-of-the-art benchmark models as well as other test models using the same signal representation in the majority of examined conditions, even when moderate amounts of noise are added to the reverberant signals.
Authors:
Santos, João F.; Falk, Tiago H.
Affiliations:
Institut National de la Recherche Scientifique, Montreal, Quebec, Canada; Centre for Interdisciplinary Research in Music Media and Technology, Montreal, Quebec, Canada(See document for exact affiliation information.)
AES Conference:
60th International Conference: DREAMS (Dereverberation and Reverberation of Audio, Music, and Speech) (January 2016)
Paper Number:
3-1
Publication Date:
January 27, 2016
Subject:
Paper Session 3
Click to purchase paper as a non-member or you can login as an AES member to see more options.
No AES members have commented on this paper yet.
To be notified of new comments on this paper you can subscribe to this RSS feed. Forum users should login to see additional options.
If you are not yet an AES member and have something important to say about this paper then we urge you to join the AES today and make your voice heard. You can join online today by clicking here.