Subject: Re: [linux-audio-dev] Gnome Wave Cleaner (audio restoration) project,request for information
From: weltyj_AT_yahoo.com
Date: Tue Mar 05 2002 - 06:37:26 EET
Paul Davis wrote:
>
> >On closer inspection, I do have a problem with buffering
> >the audio data going into OSS. Definitely my problem,
> >and easy to fix.
>
> 3 quick (but seriously intended) questions:
>
> is there a reason why your program does not use a separate program
> (e.g. play(1) or sndplay(1) or alsaplayer(1) or aplay(1)) for audio
> playback? is there a reason why your program is not written as a
> LADSPA plugin? is there a reason why your program is not written as a
> JACK client?
One of the issues I faced, was having a cursor tracking the audio
wavform display in near-realtime. I previously used COOL Edit on
windows, and the way I'd clean up audio files is by, well, listening
to them. When the you hear a click or pop, stop the playback, zoom
in on the area around the cursor, and then do some local editing. I
couldn't figure out a way to get this done easily. The OSS drivers
aren't *that* hard to program for, if you don't make silly mistakes,
like, err, I did ;-)
The click detection routine is very cpu intensive, to put it mildly.
My reference machine is my 350 Mhz pentium, and I don't suspect I'm
all that different from the "average user" in regards to cpu horsepower.
That became my frame of reference for how to design the app. My other
goal for this project was to learn programming for the gnome API, hence
my own interface.
I think everyone really wants some magical software than can just
scan an audio file, and find the noise, clicks etc, and just clean
them up. Real life is not like that (*unfortunately :). If you
are going to have a reasonable shot at cleaning out hiss, for example,
you have to identify a "silent" portion of the track. This gives you
the fft noise fingerprint that is then removed from the entire track.
Think about the songs where there is some soft violins slowly coming
up to volume, it takes a human being to know where the violins start,
and not to include that in the noise fingerprint. The easiest way
to identify a portion of the track is visually selecting it looking
at the wavform, and listening to it...
So, noise elimination requires an interactive 2 step process, first
identifying the "silent portion. Then getting the fft noise
fingerprint and applying that fingerprint to another designated
portion of the track. I don't know if plugins are capable of doing
that. I was going to start looking at doing a GLAME plugin, but
there's this other problem of my real job, and a real life ;-)
>
> it seems to me that i'd like vinyl restoration either as a regular
> unix cmdline utility (somewhat like anoi) so that i could process
> existing disk files, or as a plugin for a variety of audio processing
> networks (which is what LADSPA and JACK allow for).
>
> i don't know how CPU intensive your code is, but being able to run a
> vinyl restoration plugin in real-time within an existing audio
> processor (e.g. ardour, ecasound, glame, anything that supports
> LADSPA) strikes me as incredibly cool.
I agree wholeheartedly with that vision :-)
No way can I see it running in real time, but, with some thought
it may be possible to have a simpler approach for folks who are
looking for "quick and dirty". I can appreciate that. I find
that for tracks that are really dirty, and that I care about a lot,
I may spent 10 times the time cleaning it up than the length of the
track. That will never change, but there are probably some other
cases where an automated approach could work.
The basic denoising and declicking algorithms in GWC are reasonably
well abstracted, as well as the routines that read and write audio
data, so I don't suspect it would be a huge task to make plugins.
Any volunteers? :-)
***** PS ******
I'm having cable modem problems, might be offline for a while,
don't think I'm ignoring you!!!
>
> --p
This archive was generated by hypermail 2b28 : Tue Mar 05 2002 - 06:29:40 EET