In order to overcome a limited performance of a conventional monaural model, this letter proposes a binaural blind dereverberation model. Its learning rule is derived using a blind least-squares measure by exploiting higher-order characteristics of output components. In order to prevent an unwanted whitening of speech signal, we adopt a semi-blind approach by employing a pre-determined whitening filter. The proposed model is evaluated using several simulated conditions and the results show better speech quality than those of the monaural model. The applicability of the model to the real environment is also shown by applying to real-recorded data. Especially, the proposed model attains much improved word error rates from 13.9 ± 5.7 (%) to 4.1 ± 3.5 (%) across 13 speakers for testing in the real speech recognition experiments.
Bibliographical noteFunding Information:
The authors would like to thank the anonymous reviewers for their criticisms which improve this letter very much. Also, the authors thank Drs. Doh-Suk Kim and Hyung-Min Park for their fruitful discussions. This work was supported by the Brain Neuroinformatics Research Program sponsored by Korean Ministry of Commerce, Industry, and Energy.
- Automatic speech recognition
- Blind deconvolution
- Blind dereverberation
- Blind least squares
- Independent component analysis
- Speech enhancement
ASJC Scopus subject areas
- Computer Science Applications
- Cognitive Neuroscience
- Artificial Intelligence