Sound cards (Was: Re: no sound

Hartmut Noack zettberlin at
Fri May 20 14:53:29 UTC 2011

Am 20.05.2011 15:15, schrieb Ralf Mardorf:
> On Fri, 2011-05-20 at 15:00 +0200, Hartmut Noack wrote:
>> Am 20.05.2011 14:37, schrieb Ralf Mardorf:
>>> On Fri, 2011-05-20 at 14:04 +0200, Thomas Orgis wrote:
>>>> Am Fri, 20 May 2011 13:54:57 +0200
>>>> schrieb Ralf Mardorf<ralf.mardorf at>:
>>>>> When recording soft synth just by
>>>>> JACK, without the sound cards being involved, there's a loss for the
>>>>> sound quality too!
>>>> Wait a minute... could you explain that? You have a loss of quality compared to live playback of the soft synths (using JACK?) when playing back a recording taken from JACK? A recording that preserves 32 bit floating point sample format (heck, or 24 bit integer) and the sample rate, of course?
>>> Yes and other people who can't hear it, do have it too.
>> I do not.
>>> You can see it
>>> by watching the waves spectral by Audacity. I did this regarding to a
>>> zero-copy issue, that appears if a Jack client is connected directly to
>>> itself, e.g. to do the mastering. 48 and 96 KHz, 32-bit wav 32-bit
>>> float.
>> If a synth has dynamic filters it will never produce the exactly same
>> stream twice. But if you think about yourself you will find out, that
>> given you use the same settings for Jack on a HDA or a HDSP you will get
>> exactly the same quality.
> I'm an expert for audio engineering. I did work for Brauner microphones
> development and others, hence I know a little bit about how to do
> tests ;).
> No dynamic filters are involved!
> It's very simple, there's a natural sounding drum set as example drum
> kit for Hydrogen. Play a rhythm, record this Rhythm and then record this
> recording and compare both recordings. They should be equal, but they
> aren't equal. I can here a !clear! loss and it's visible by spectral
> waves.
>> Simply because a synth-software only delivers, what it renders to Jack
>> and Jack does *not* change anything in that rendered data. There is
>> simply not soundcard and not even a driver involved in the rendering
>> itself. DSPs only do the very same thing faster as cheap chips.
>> All difference in sound quality is related to DAC/ADC period
> No! Before any converter is involved, there at least could be rounding
> errors, if you don't use 32-bit float all the times.

And why should I not use 32bit float all the time?

Of course there are differences, if format-conversion is involved. But 
you did not mention such conversions, you only talked about sound cards 
causing mysterious differences when Jack delivers a stream from a 
synth-application directly to a recorder.

The normal, sane setup fpr recording a synth directly with Jack is, that 
the synth, Jack and the recorder all run with the same samplerate and 
32bit float or at least all 3 with 16bit Int. And if that is set up like 
this, there is zero influence of the soundcard on the recording.

> And by the way, the sound card will effect the original and the digital
> copy in the same way, even with a bad sound card both recordings
> shouldn't differ.
> Hey, do a recording of a recording

You mean, like recording something from ams via jack then play the 
wav-file with mhw and record this with ardour. Then compare the two 
recorded streams?

If in such a process recording A would differ from recording B then MHW 
or Ardour *could* cause such a difference. Jack itself could only be 
charged, if resampling and/or dithering would be involved. That is: if 
the synth-engine would work with 44.1Hz while Jack runs with 48Hz. And 
you will not want to set up a synth like this.

But you said, that there would be a difference between a 1st-level 
recording from a synth via Jack with sound card A and another recording 
of the same synth with soundcard B.

And that is not the case.

> and then run the diff command to
> compare them ;)!
>>>> I have to wonder what you did there to alter the data from the soft synth.

Format conversion?
Some synths (like ZynaddSubFX) can be configured to produce streams in a 
certain format (such as 44.1 Hz / 16bit Int) and still run with Jack 
that runs with different settings.
Every body sets Zynadd to run with the same SR as Jack and Zynadd 
recommends that if started from the command line.

>>> I mean ... we're talking bit-exact copy here, aren't we? Can you present a test setup to observe that issue?
>>> Any Linux install I know, e.g. 64 Studio 64-bit 3.0, 3.3, Suse 64-bit
>>> 11.2 and Edubuntu 32-bit Maverick + Ubuntu Studio meta packages and
>>> others! If you can't here it, try to see it. If you don't have this
>>> issue too, some people claim that they get 100% correct digital copies,
>>> then something on my machine might cause a software issue, but I don't
>>> think so.
>>> Ralf
>>>> Alrighty then,
>>>> Thomas.

More information about the Ubuntu-Studio-users mailing list