[pulseaudio-discuss] Microphone DC-Offset compensation and noise filtering.

IL'dar AKHmetgaleev akhilman at gmail.com
Tue Jan 11 01:53:47 PST 2011


На Mon, 10 Jan 2011 11:44:36 +0100
Maarten Bosmans <mkbosmans at gmail.com> записано:

> 2011/1/10 IL'dar AKHmetgaleev <akhilman at gmail.com>:
> > I have a cheap USB headset which records audio with the little
> > DC-Offset and noise at 50Hz.
> >
> > As I know it's very usual problem of cheap audio devices and
> > integrated audio cards.
> >
> > So I'm requesting a module which will filter input stream.
> > This pseudo code was suggested in alsa ML:
> >
> >> xs=0
> >> while (input){
> >> xs=.01 xinput +.99 xs
> >> xoutput=xinput-xs
> >> }
> >> (This averages over roughly the last 100 inputs and subtracts the
> >> offset). If you want a longer averaging, change the coefficients.
> 
> It is wrong to say that the averaging is done over the last 100
> inputs. What you are proposing is averaging with exponential
> weighting. This means the averaging window is infinately long.
> 
> The concept can be useful, but you need a much smaller factor if by
> inputs you mean samples. 0.00001 for example would be better, because
> then the last period of a 20Hz wave in a 48000Hz signal only
> contributes 2.4% to the average (1-.99999^2400), which seems much more
> reasonable.
> 
> > Will be nice to have such module with sampling period as attribute.
> 
> Is there any reason this kind of signal processing wouldn't be more
> appropriate in an alsa driver?

I had tested ladspa filtering and sound becomes very nice.
I did it with alsa ladspa plugin and with this chain:
  source->Eq(cutting off all behind 100 Hz)->
     se4(to decrease dynamic range)->amp_mono->
     dysonCompress->recorder
But this alsa plugin makes pulseaudio hangs with some applications
(skype and teamspeak3).

I had tried same using PA chain:
  source->loopback->multiple ladspa-sinks->null-sink|monitor->recorder.
But because of Eq has different channel naming I had to use another
equalizer. With this chain delay was much bigger then with alsa
approach.

Will be nice to have ladspa-source and null-source modules also.



More information about the pulseaudio-discuss mailing list