[ 3 / biz / cgl / ck / diy / fa / ic / jp / lit / sci / vr / vt ] [ index / top / reports ] [ become a patron ] [ status ]
2023-11: Warosu is now out of extended maintenance.

/sci/ - Science & Math

View post   
File: 318 KB, 634x875, 1463683265485.jpg [View same] [iqdb] [saucenao] [google]
8087091 No.8087091 [Reply] [Original]

Couldn't we shrink the size of an audio file without impacting the actual sound too much by taking the Fourier transform and eliminating frequencies with low amplitudes?

>> No.8087093

>>8087091
We already lower the file size by eliminating frequencies humans cannot hear.

>> No.8087095

>>8087093
When they do this, do they take the dft of the whole long audio signal at once or do they split it into smaller chunks and then recombine?

>> No.8087100

>>8087095
it depends on the encoding format.
you can do that if you want.

if you want to know more about how mp3 works, read this http://www.mp3-tech.org/programmer/docs/mp3_theory.pdf

too bad you posted with a gorilla, I almost auto-hid your thread.

>> No.8087102

>>8087100
Thanks dude, we've been using the Fourier series in my analysis course and this application possibility occurred to me

>> No.8087124

>>8087102
it's a good thing it did, we need more people like you.

I'm just trying to think about your actual question. I know that for images, the format jpg divides the image (which is basically a 2D signal) into small squares and compresses each square.


Ok I got it:

ideally, you would take the whole signal and try to compress it as you suggested.

HOWEVER: if you use 4 n*n squares, it roughly takes 4* n^4 computations. If you use one 2n*2n square, it will take 16n^4 computations.

So it's more computationnally advantageous to take small pieces of an image and compress them. But if the pieces are too small, you get artifacts. So in jpg, the choice of using 8*8 squares of pixels is a tradeoff.

Now sound is a 1D signal. If you take a signal of length 2n, it would take 4n^2 computations. If you take two signals of length n, it would take 2*n^2 computations. It's still better to divide the signal. But it's not as important as for images.

I'm guessing older encoding methods divide the sound file into small pieces before compressing, but newer encoding methods probably take the whole signal.

>> No.8087197

>>8087091
Fuck off Gorilla-shitter.

>> No.8087522

>>8087124
Interesting, thanks