IPB

Welcome Guest ( Log In | Register )

Editing A Wave With A Program..., How would I do it?
Mac
post Nov 4 2002, 20:43
Post #1





Group: Members
Posts: 650
Joined: 28-July 02
From: B'ham UK
Member No.: 2828



I have some wave files from a set of cd's that I'm trying to fix. They have what I'd call "digital screw-up" on them (these are purchased cd's, yet have artifacts like a bad cd-r)..

It appears that groups of 5 samples in the right channel are offset by varying amounts, in patches throughout the tracks. It sounds incredibly annoying, and I have tried lots of different ways to fix them, filtering, click removal, interlacing other sections of the song over them. They definately improve the sound quality, but naturally, I want what they *originally* sounded like!

What I dawned on recently though was that the affected samples, and their offsets seem to follow a pattern. You get regularly spaced groups of 5, with regular spacing between each bad sample. And the amount it is offset by, from what I can tell so far, it appears to be by the difference between it's actual value and the value before it.. which would be hard to work out mathematically, because how are you meant to know what it "should" be?

Anyway, the best way I thought to fix this problem would be to get a friend to write a program that manipulated the samples in mathematical form. I can do maths, so could probably find a reasonable way of correcting the samples, then he could write a program that performed the algorithm on the correct samples in a wave.

My question is, could this be done by converting a wave file into a file that contains the sample values as integers? Or is this what a wave file is, meaning it could be acted on directly by a program that could search through the values and alter certain ones?

Thanks! smile.gif

(i can post samples for any kind person who wants to help figure out how to fix them!)


edit:
Ok, I've looked at the files some more, and noticed something quite obvious. The problem samples all take the same value as their preceding sample. I found that taking the difference between the corresponding left channel's sample and it's preceeding sample was a good approxomation (it moved the bad samples in the right direction and close to the right amount), and that would be why.

Is there a way of predicting how the samples should be changed to follow their natrual curve (as would be hinted at by the preceeding and following samples?) It would need to be specifically for when one sample is of unknown value, but the others are correct.. I'm not sure if standard click-restoration in Cool-Ed can deal with this.. just single bad samples, rather than a click in a vinyl which spans dozens of samples...

This post has been edited by Mac: Nov 4 2002, 21:08


--------------------
< w o g o n e . c o m / l o l >
Go to the top of the page
+Quote Post
 
Start new topic
Replies
Mac
post Dec 7 2002, 12:56
Post #2





Group: Members
Posts: 650
Joined: 28-July 02
From: B'ham UK
Member No.: 2828



I'll make a separate posts for the results and method... smile.gif


I've taken ErikS's advice on this and have used Matlab. It's an awesome program biggrin.gif Thanks smile.gif


An input .au file can be read and converted into a table of numbers. We then scanned through this list and picked out values that were the same as the previous value. Discarded ones that weren't in a position divisible by 16 (to exclude any samples that were meant to be like that due to the quantization of 16 bit)..

We then removed all these bad samples from the dataset, then used cubic-spline fitting to estimate these missing values. The resulting file could be saved as .au and compared.

I know spectral graphs aren't the right way to go about it, but check this :

http://www.mwstore.co.uk/crap/DnB/x310_1o.jpg

The bottom is the *before* shot, the top is the *after*. I think that's an amazing improvement!!!


If you wish to hear for yourself, and analyse the broken file more closely, here they both are (273kb each):

Broken sample
Fixed sample



I have learnt a lot about program strcutres and Matlab doing this! I'm definately gonna buy the dude who helped me a beer! He even likes drum'n'bass too (that's what some these samlpes are of).



I'm going to try and improve on the results, as I have the L-R channel difference as well... I can use the same matrix of broken sample locations and apply it to this 2nd file, which usually contains a much smoother waveform, and so will be predicted better I hope biggrin.gif

Enjoy!


--------------------
< w o g o n e . c o m / l o l >
Go to the top of the page
+Quote Post

Posts in this topic


Reply to this topicStart new topic
1 User(s) are reading this topic (1 Guests and 0 Anonymous Users)
0 Members:

 



RSS Lo-Fi Version Time is now: 16th April 2014 - 12:06