Live Q&A - Speech Dereverberation using Coherent to Diffuse Power Ratio Estimators (CDR)
David Castro Pinol - Watch Now - Duration: 22:32
15:10:32 From Brewster LaMacchia : (I have to drop but thank you for putting together this presentation) 15:14:39 From JohnP : I'm interested as to why David would not apply these general principles to electromagnetic propagation ? 15:15:18 From Thomas Wagner : In the postprocessing of the voice audio, it sounded like there were some high frequency residuals. Is that inherent to the algorithm or is it something that could be improved about it? 15:16:20 From Thomas Wagner : @JohnP a hunch would be processing speed but that's the same thing isn't it? 15:18:12 From Michael Kirkhart : Yes, I noticed that as well. 15:18:19 From JohnP : @Thomas processing speed is just a technology issue 15:19:48 From Thomas Wagner : @JohnP yes sure is and not so much of a problem these days. I just meant to say I couldn't see the difference. 15:20:17 From David Castro Piñol : https://github.com/andreas12345/cdr-dereverb 15:20:36 From Thomas Wagner : Thank you David for your answer. Therefore is it safe to assume you could integrate some ML /DL in this algorithms to get the optimal performance out of it? 15:22:24 From Thomas Wagner : Or even the parameters you were talking about but I didn't think about working with CBR directly. 15:22:41 From Thomas Wagner : That's a good point thank you 15:23:56 From Michael Kirkhart : Updated code link works - thank you! 15:24:29 From Michael Kirkhart : Link to paper (from github repo): https://arxiv.org/pdf/1502.03784.pdf 15:26:00 From Thomas Wagner : Thank you bye!