- (2018.08) Our paper "All-neural online source separation, counting, and diarization for meeting analysis'' was accepted for ICASSP-2019. You can check out our demo to see how it works.
- (2017.08) We gave a presentation at ICASSP-2018 about "Listening to each speaker one by one with reccurent selective hearing networks.'' It's a DNN framework to jointly perform source separation and source counting.
- (2017.08) We gave a presentation at Interspeech-2017 about "Neural Network-Based Spectrum Estimation for Online WPE Dereverberation.'' It is an DNN-based extention of state-of-the-art speech dereverberation called weighted prediction error (WPE).
- (2017.03) We gave a presentation at ICASSP-2017 about "Deep mixture density network for statistical model-based feature enhancement.''
- (2015.12) We won the CHiME-3 challenge. See the CHiME-3 challenge webpage for mote details.
- (2014.05) We organized REVERB challenge in collaboration with E. Habets, R. Haeb-Umbach, V. Leutnant, A. Sehr, W. Kellermann, R. Maas, S. Gannot and B. Raj. Many research institutes (including us, NTT) took part in the challenge to evaluate their own dereverberation algorithm and ASR technique based on a common evaluation database.
See the challenge webpage for mote details.
- (2011.09) We won the CHiME-3 challenge. Demo sounds for denoising method (highly non-stationary noisy environment) presented at CHiME challenge were uploaded here.
- (2009.02) Developed a commercial dereverberation software (Pro Tools plugin). Demo sounds and a demo version of the software are now available at this webpage.