Clipfix (see waveform pic) - plugins available?

[steve]

I am also using audacity-1.3.5-beta (unicode) on ubuntu-8.10.
Apart from the addition of a few Nyquist and LADSPA plug-ins, I have a standard installation of Audacity 1.3.5 installed with Synaptic.

My computer is a 500MHz Pentium III with 512 MB of RAM.

I have just tested the ClipFix plug-in (the ZIP file downloaded from Gale's link) on 130 second recording at 44100Hz 16 bit mono. Processing was rather slow (took about 10 minutes) but it completed successfully and repaired the clipped audio.

So what have I done different to you?

[scar]

i have found some sort of solution which satisfies me, but it took me about 1 hour to complete on that 80 minute track. also, audacity crashed several times during this process. i would of course like clipfix to work with one pass, which would only take me a few seconds to run, and the computer a few minutes. so here's what i did, would be nice to have this automated perhaps.....

1.) ran "find clipping"
now i have a separate track that shows where the clipping is and do not need to rely on the red lines.
2.) amplified by -10 dB
3.) zoomed into each area marked by the clipping track
4.) selected a few tenths or hundredths of seconds around where the clipping occurred
5.) ran clipfix at 95% on the selection
6.) zoomed out and moved on to next clipped area, repeated until all clipped areas had been fixed

[scar]

I am also using audacity-1.3.5-beta (unicode) on ubuntu-8.10.
Apart from the addition of a few Nyquist and LADSPA plug-ins, I have a standard installation of Audacity 1.3.5 installed with Synaptic.

My computer is a 500MHz Pentium III with 512 MB of RAM.

I have just tested the ClipFix plug-in (the ZIP file downloaded from Gale's link) on 130 second recording at 44100Hz 16 bit mono. Processing was rather slow (took about 10 minutes) but it completed successfully and repaired the clipped audio.

So what have I done different to you?

maybe because it's mono, and mine is 32-bit float stereo?

i have amd64 so maybe that is different?

also, i am using the plugin from,

http://audacity.cvs.sourceforge.net/vie ... ortby=date

[steve]

OK, I've tried it on a short 32 bit (float) stereo file and it still works.
Perhaps you should try replacing your version of ClipFix with the one in Gale's link.

[Edit] I've had a quick look and I've not spotted the differences yet, but the two versions are not identical.

[scar]

can you provide the link again, just to be clear? thx

[Gale Andrews]

My copy of ClipFix has now been updated to be the same as in source code, but the only difference between that and what was there before is the ;categories line which means that 1.3.6 sorts it under Effect > Utility > Noise Removal.

One of our contributors who works on Nyquist plug-ins has come back "on stream" again recently, I'll ask him if there is some way to improve performance. Reducing "largenumber" (the maximum number of samples it works with) would be, but it doesn't then carry on after it's dealt with that number, but stops and truncates the audio to that length.


Gale

[scar]

thanks. so am i correct in understanding that the truncation of the whole selection is normal, for now? but then i do not understand how stevethefiddle is able to successfully execute the plugin on a long sample of 130 seconds, when i have tried using it on as little as 1 second and experience deletion or truncation.

is it possible to modify/update this plugin so it can be successfully executed in one pass on a huge selection, like 80 minutes, like the audio screenshot i posted earlier? or is that just a pipe dream?

[steve]

I have attached the version of ClipFix that I am using. Remove or rename your existing clipfix.ny file, then unzip this file into your Audacity Plugins folder then restart Audacity.

There is no guarantee that this is the problem - in fact I suspect that it isn't the problem, but it's worth trying as it will at least rule out one possibility.

The way that clipfix.ny is written, it reads the audio into memory in order to be able to do certain operations on it, such as finding peak levels. When it does this, it requires about 4 bytes per sample. The program uses several functions which require that the sample is loaded into RAM, so removing the limitation is more than a small change in the code.

The version that I am using limits the number of samples that can be processed to "largenumber" - I don't remember off the top of my head how many samples that is - perhaps a million. I think that there may have been an earlier version that limited the number of samples to "NY:ALL", which is a billion samples and would require about 4GB of available RAM.

Reading an 80 minute file into RAM would require getting on for a Gig of RAM, so it would be possible, but would crash on computers that had insufficient RAM.

[Gale Andrews]

I have attached the version of ClipFix that I am using.

This is the same as the one on my site and in source code. The "largenumber" here is 100 million samples, so at 44100 Hz sample rate it will truncate audio longer than (100000000/44100) / 60 =37.792895 minutes.


Gale

[Arnie]

Gale was kind enough to notify the mailing list that this thread has been revived. I'm sorry for not getting back at you earlier, my studies have been quite demanding. (It's my excuse for the next 5 years.) But I'm glad that you guys are working on it, your work is really appreciated.

I see that the topic has shifted over to the performance of the clipfix plugin. Is there any chance you will look into the inverse polarity clipping (see OP) as well? I actually gave up fixing my messed up recording and I took down the sample Wave (page 2) in a cleanup of my webspace, but if there's any interest I'd be glad to put it back. The last post on page 1 by stevethefiddle outlines (in pseudocode) how the polarity fixer could be implemented in clipfix.ny.

As for the perfomance: it doesn't really bother me because personally I just run Clipfix on small selections in a recording instead of the entire wavefile. Am I doing it wrong? My thought is that if a recording is clipping all over it's usually so messed up there's no need trying to salvage it anyway.

Wishing you a merry Christmas,

Arnoud