Laurent Duval provides a small summary of the talk by Albert Cohen on Compressed Sensing yesterday in the continuing series at IHP.
Steve Eddins at Matworks (the maker of Matlab) talks about Doug Williams who draws a parallel between the innovation brought forth by the Faster Fourier Transform in the West (FFTW) and Compressive Sampling in Reconsidering the impossible. We have indeed noted that the "impossible" word was uttered by reviewers when providing comments on compressed sensing papers.
At ICASSP there was a presentation that seemed to have produced some high-impact TechMeme style buzz : Music File Compressed 1,000 Times Smaller Than Mp3. I could not find a paper on the topic yet but if I understand correctly, they produced additional compression though some knowledge of the Clarinet instrument as opposed to blindly guess some priors on the sound as is done on the MP3 format. In terms of Compressed Sensing, it looks like they clearly have a model for clarinet that could be equated to a small sized dictionary specific for clarinet that enables them to deal with the small manifold of clarinet sounds: This finding maybe the beginning of a larger effort in building a Compressive Sensing Microphone. As pointed out before, the building of such hardware would relegate to an afterthought the finding of the actual dictionary (be it a family of wavelet or a model specific clarinet dictionary).
Credit: NASA/JPL/Space Science Institute, Janus: a Moon of Saturn photographed on Feb. 20, 2008.
Steve Eddins at Matworks (the maker of Matlab) talks about Doug Williams who draws a parallel between the innovation brought forth by the Faster Fourier Transform in the West (FFTW) and Compressive Sampling in Reconsidering the impossible. We have indeed noted that the "impossible" word was uttered by reviewers when providing comments on compressed sensing papers.
At ICASSP there was a presentation that seemed to have produced some high-impact TechMeme style buzz : Music File Compressed 1,000 Times Smaller Than Mp3. I could not find a paper on the topic yet but if I understand correctly, they produced additional compression though some knowledge of the Clarinet instrument as opposed to blindly guess some priors on the sound as is done on the MP3 format. In terms of Compressed Sensing, it looks like they clearly have a model for clarinet that could be equated to a small sized dictionary specific for clarinet that enables them to deal with the small manifold of clarinet sounds: This finding maybe the beginning of a larger effort in building a Compressive Sensing Microphone. As pointed out before, the building of such hardware would relegate to an afterthought the finding of the actual dictionary (be it a family of wavelet or a model specific clarinet dictionary).
Credit: NASA/JPL/Space Science Institute, Janus: a Moon of Saturn photographed on Feb. 20, 2008.
It was Doug Williams who drew the parallel between FFTW and compressive sampling. I was simply quoting from his editorial in the most recent issue of IEEE Signal Processing Magazine.
ReplyDeleteCorrected.
ReplyDeleteThanks Steve.
Igor.