FFT, now even faster

[benb]

The faster-than-fast Fourier transform

For a large range of practically useful cases, MIT researchers find a way to increase the speed of one of the most important algorithms in the information sciences.

http://web.mit.edu/newsoffice/2012/faster-fourier-transforms-0118.html

... but it's still not faster than the speed of light.

 

[Pano]

Hmmmm.......

 

[DF96]

MIT is full of clever people, so we can assume the developers of this method are smarter than the PR man who wrote the linked piece. He seems to think that chamber music is 'sparse' - maybe sparser than a full orchestra, but I wouldn't want to hear a quartet replaced by a few gated sine waves!

 

[jcx]

well the practical lesson from perceptual audio compression is that you can get quite "transparent" music reproduction from <20% of RedBook Shannon-Hartley Channel Capacity bit rate - so "sparse" could be a fair characterization for music

the compression algorithms currently used in audio have a harder time with the broadband noise of the audience applause

 

[DF96]

I remember some years ago listening to Classic FM on DAB (MP2 coding at 160 kbit/s). The music was tolerable, but the applause at the end sounded just like someone varying the level of a white noise generator. Horrible! I switched back to FM and that was fine. Not long after that I stopped listening to DAB.

 

[CopperTop]

I often hear MP3-style compression in cymbals; stuff that is noise-like, but not actually noise, and your brain can spot the structure even if the compression algorithm can't. You can also see the same thing in MPEG'ed video, where 'noise-like' scenes reveal the edges of the DCT-ed blocks. I particularly notice it on crowd scenes and the surface of rough water.

 

[DF96]

A camera pan during football makes the blades of grass disappear on digital TV - looks like a billiard table instead. Once the camera stops moving the grass reappears.

 

[a.wayne]

When ... during the WC ? Too much Stout + England's drubbing ....

 

[benb]

MIT is full of clever people, so we can assume the developers of this method are smarter than the PR man who wrote the linked piece. He seems to think that chamber music is 'sparse' - maybe sparser than a full orchestra, but I wouldn't want to hear a quartet replaced by a few gated sine waves!

well the practical lesson from perceptual audio compression is that you can get quite "transparent" music reproduction from <20% of RedBook Shannon-Hartley Channel Capacity bit rate - so "sparse" could be a fair characterization for music

the compression algorithms currently used in audio have a harder time with the broadband noise of the audience applause

Yeah, reading the article, it seems like it's the same old MP3 type encoding being used for more things. The writer does seem to think a string quartet is just four sine waves.

The more I read, the less "new" this sounds. The Goertzel Algorithm is just a digital filter calculation to do a single FFT bin, and this talks about doing fewer FFT binss, but you still have to do a full FFT at the start to know what bins have "almost" no signal so they can be (possibly) ignored.

As I've seen pointed out before, with "digital" we don't get The Miracle Of No Distortion, we get different and never-before-heard distortions.

 

[ontoaba]

I often hear MP3-style compression in cymbals; stuff that is noise-like, but not actually noise, and your brain can spot the structure even if the compression algorithm can't. You can also see the same thing in MPEG'ed video, where 'noise-like' scenes reveal the edges of the DCT-ed blocks. I particularly notice it on crowd scenes and the surface of rough water.

128kBps MP3 is not CD quality, I could still clearly easy to hear the differences with 192kBps version. If I converting CDs, I use 320kBps, space isn't problem for this days storage system. Compression logarithm couldn't keep its fast changes properly.