21 Nov
2014
21 Nov
'14
12:42 p.m.
I may perspectively run into a shortage of audio inputs and stumbled
across a blog, that reads like it is possbile to "merge" two soundcards
into one virtual, that eventually looks as a single one to alsa.
However, that blog does not mention jack.
Also, in an stone old ardour thread Paul Davis suggests to a similar
problem: ", then you simply get 2 digifaces", but again, this may have
been before the rise of jack.
And now I am wondering wether this is a possible or even reliable way to
trick jack into being able to to finally handle more than one soundcard,
if presented by alsa as being a single one.
I am honest, that asoundrc is way beyond me, as alsa alway has been,
thats why I am asking and I am thinking about using two ordinary, 1st
generation, pre HDSP, Hammerfall cards to expand ADAT IO. Would be
cheaper than investing into a raydat, when time has finally come.
The posts for reference:
http://soundofsyntax.wordpress.com/2012/02/12/how-to-configure-ubuntu-11-10…
https://community.ardour.org/node/141
Any thoughts on this? Or won't it work with jack?
21 Nov
21 Nov
1:56 p.m.
On Thu, Nov 20, 2014 at 3:56 PM, Ede Wolf <listac(a)nebelschwaden.de> wrote:
I may perspectively run into a shortage of audio
inputs and stumbled
across a blog, that reads like it is possbile to "merge" two soundcards
into one virtual, that eventually looks as a single one to alsa. However,
that blog does not mention jack.
Also, in an stone old ardour thread Paul Davis suggests to a similar
problem: ", then you simply get 2 digifaces", but again, this may have been
before the rise of jack.
And now I am wondering wether this is a possible or even reliable way to
trick jack into being able to to finally handle more than one soundcard, if
presented by alsa as being a single one.
this is a function of the ALSA driver for a given device, plus the
capabilities of the device itself. We still don't know for sure whether
ALSA can get all the cards started reliabily in < 1sample, but it seems as
if it should.
a more device independent way of doing this, but one that implies sample
rate conversion going on, is to use zita-a2j to allow JACK to use more than
one device. With current JACK1, this is even builtin to JACK itself and can
be done from the command line.
3:11 p.m.
On Thu, Nov 20, 2014 at 04:33:57PM -0500, Paul Davis wrote:
a more device independent way of doing this, but one
that implies sample
rate conversion going on, is to use zita-a2j to allow JACK to use more than
one device. With current JACK1, this is even builtin to JACK itself and can
be done from the command line.
Another way is to use the soundcard of a a second computer and link
the two using zita-njbridge. If the connection is via a LAN and
Jack on the second PC is only lightly loaded you can actually
achieve the same latency as with zita-ajbridge. This is so because
njbridge assumes the worst case: the transmitter running near the
end of its Jack cycle. If that is not the case you can set the
extra buffering in zita-n2j to zero and still have some headroom
for network delay.
At my new workplace (*) I've ported njbrigde to OSX and Windows.
So we can now exchange full quality multichannel audio with low
latency between Linux, OSX and Windows machines in any combination.
(*) Since some people have asked: I'm now in Munich and working
on 3D audio at the European Research Center of Huawei.
Schuss,
--
FA
A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)
3:12 p.m.
Thanks very much. I do know about zita-a2j any may end up using it for
recording smtpe through onboard audio, at least as an experiment, but
for standard audio I would like to avoid resampling whenever possible.
Also, maybe pure superstition, that a2j attached soundcard always feels
as just being second best, not treated equal.
However, if I may re-ask, as it has not been quite clear to me: While
alsa seems to be capable of handling two cards that way, would this also
work for jack or would jack refuse to fail for this trick?
Assuming, that jack sits on top of alsa, it should not care about
whatever witchcraft alsa brews beneath. But what do I know?
So is there a realistic chance that I would be presented with 52 inputs
instead of 26 as of now? Without using a2j?
For synchronizing both cards I'd use external wordclock.
Am 20.11.2014 um 22:33 schrieb Paul Davis:
On Thu, Nov 20, 2014 at 3:56 PM, Ede Wolf <listac(a)nebelschwaden.de
<mailto:listac@nebelschwaden.de>> wrote:
I may perspectively run into a shortage of audio inputs and stumbled
across a blog, that reads like it is possbile to "merge" two
soundcards into one virtual, that eventually looks as a single one
to alsa. However, that blog does not mention jack.
Also, in an stone old ardour thread Paul Davis suggests to a similar
problem: ", then you simply get 2 digifaces", but again, this may
have been before the rise of jack.
And now I am wondering wether this is a possible or even reliable
way to trick jack into being able to to finally handle more than one
soundcard, if presented by alsa as being a single one.
this is a function of the ALSA driver for a given device, plus the
capabilities of the device itself. We still don't know for sure whether
ALSA can get all the cards started reliabily in < 1sample, but it seems
as if it should.
a more device independent way of doing this, but one that implies sample
rate conversion going on, is to use zita-a2j to allow JACK to use more
than one device. With current JACK1, this is even builtin to JACK itself
and can be done from the command line.
3:42 p.m.
On Thu, Nov 20, 2014 at 11:12:09PM +0100, Ede Wolf wrote:
Thanks very much. I do know about zita-a2j any may end
up using it
for recording smtpe through onboard audio, at least as an
experiment, but for standard audio I would like to avoid resampling
whenever possible. Also, maybe pure superstition, that a2j attached
soundcard always feels as just being second best, not treated equal.
You say 'feels' and that's what it is. There is no objective reason
why resampling would reduce sound quality provided it's done as it
should be done (which is the case for ajbridge). Basically the process
is the same as the anti-aliasing filter of an AD or DA converter.
And almost all such converters include digital resampling as part of
the process.
So is there a realistic chance that I would be
presented with 52
inputs instead of 26 as of now? Without using a2j?
For synchronizing both cards I'd use external wordclock.
The next release of ajbridge will have the option to bypass
the resampling in that case. The only remaining effect would
be a somewhat higher latency for the second card (because the
periods are not synchronised).
Ciao,
--
FA
A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)
22 Nov
22 Nov
2:32 p.m.
The only remaining effect would
be a somewhat higher latency for the second card
Which actually even _does_ make it second best? Nevertheless I do
believe you, that this may be negligible, however, being honest, "It is
harder to crack prejudice than an atom." (tm) That just holds true for
me in this regard, I am afraid.
Of course thanks for your (both) insight! Highly appreciated and maybe
one day I'll learn.
3:04 p.m.
On 21/11/14 20:57, Ede Wolf wrote:
merging sound cards means either you can run both off the same clock, or one is
being re-sampled. Some soundcards can run using external clocks ... so with
appropriate cabling or settings very specific to your setup it is possible to
avoid resampling ... but most merging-into-one will involve resampling, with or
without advertising that this is the case, because it works well.
Simon
The only remaining effect would
be a somewhat higher latency for the second card
Which actually even _does_ make it second best? Nevertheless I do believe you,
that this may be negligible, however, being honest, "It is harder to crack
prejudice than an atom." (tm) That just holds true for me in this regard, I am
afraid.
8:50 p.m.
On Fri, Nov 21, 2014 at 5:14 AM, Simon Wise <simonzwise(a)gmail.com> wrote:
merging sound cards means either you can run both off
the same clock, or
one is being re-sampled. Some soundcards can run using external clocks ...
so with appropriate cabling or settings very specific to your setup it is
possible to avoid resampling ... but most merging-into-one will involve
resampling, with or without advertising that this is the case, because it
works well.
this still doesn't address the synced-start question, which it would be
really nice to have a definitive answer on, at least for specific cards.
hint, hint LAU-universe.
11:15 p.m.
New subject: questions about resampling - was - jack and the merging of soundcards
On Fri, 21 Nov 2014, Simon Wise wrote:
merging sound cards means either you can run both off
the same clock, or one
is being re-sampled. Some soundcards can run using external clocks ... so
with appropriate cabling or settings very specific to your setup it is
possible to avoid resampling ... but most merging-into-one will involve
resampling, with or without advertising that this is the case, because it
works well.
Ok, so open question then. Does it make a difference to quality what the
original rate is? For example: the direct card with no resample is a 48k,
the second card will be resampled, is it better to be as close as possible
to 48k or would starting at 96 (or whatever) work better?
In all this I would assume, stereo pair inputs should be kept to one card.
Or a suround group for that matter. Or does the internal resample for
filtering make that not an issue anyway?
One more... if a group of inputs that start synced are resampled, do they
have a better post resample channel to channel sync that accross two
intefaces with different sample rates before resampling? (this is sort of
what the above questions are getting at though there amy be other issues
as well)
In the end, most people are comparing these questions to analog with the
idea that there is no phase shift in analog... I can see that two analog
channels through two eqs with the same circuit, set the same way, will
still likely have some phase shift between them. So is the channel
differences caused by digital manipulation better than analog? The same?
Worse?
--
Len Ovens
www.ovenwerks.net
23 Nov
23 Nov
7:06 a.m.
New subject: questions about resampling - was - jack and the merging of soundcards
On Fri, Nov 21, 2014 at 06:17:43AM -0800, Len Ovens wrote:
Ok, so open question then. Does it make a difference
to quality what
the original rate is? For example: the direct card with no resample
is a 48k, the second card will be resampled, is it better to be as
close as possible to 48k or would starting at 96 (or whatever) work
better?
First, assuming the resampling is done correctly, there is no loss
of 'quality' in the sense of distortion, added jitter, etc. etc.
The only differency will be some change in frequency response, and
then only for frequencies close the half the sample rate.
In practical cases, if all sample rates are at leat 44.1 kHz, the
differences are so tiny that they become completely academic.
In all this I would assume, stereo pair inputs should
be kept to one
card. Or a suround group for that matter. Or does the internal
resample for filtering make that not an issue anyway?
Any digital resampling done by a single device or software resampler
will be identical on all channels, so it doesn't change the relative
phases of the channels. So yes, stereo pairs, AMB signals etc. should
be on the same device.
One more... if a group of inputs that start synced are
resampled, do
they have a better post resample channel to channel sync that
accross two intefaces with different sample rates before resampling?
(this is sort of what the above questions are getting at though
there amy be other issues as well)
I don't understand the question... could you rephrase it ?
In the end, most people are comparing these questions
to analog with
the idea that theis no phase shift in analog... I can see that
two analog channels through two eqs with the same circuit, set the
same way, will still likely have some phase shift between them.
Maybe, but if the amplitude responses are the same the phase responses
will be the same as well. And even if the amplitude responses are not
exactly the same but still similar, the differences in phase response
will be very small. This is so because for the simple circuits used
for audio EQ the amplitude and phase responses are linked by mathematical
relations. It is actually quite difficult to control them separately
with analog circuits. Whatever phase differences there will be between
'identical' EQs will be very small and of no consequence. One example:
if the phase difference between two equal amplitude signals is 30
degrees, then the sum will be 0.3 dB down. That will be a tiny fraction
of the amount of EQ you are applying anyway.
So is the channel differences caused by digital
manipulation better
than analog? The same? Worse?
Unless you do something very weird, there will be no differences
in the digital case.
Ciao,
--
FA
A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)
22 Nov
22 Nov
5:15 a.m.
New subject: questions about resampling - was - jack and the merging of soundcards
On Sun, 23 Nov 2014, Fons Adriaensen wrote:
On Fri, Nov 21, 2014 at 06:17:43AM -0800, Len Ovens
wrote:
You answered it when you answered the other questions. Basically, It
appears there are still some artifacts between devices if they are not
sample synced at the codec. While not enough to hear alone, perhaps enough
to hear as part of a stereo pair. This seems to agree with some of the
multicast papers that suggest sending only a unified group of channels
(stereo or suround) in one stream.
So I was asking if the audio phase/sync would be preserved if a stereo
pair was split across two devices, one out of sync with the other and
resampling being used to sync the two together. It appears from your other
answers, that there may be noticable artifacts.
My initial thoughts on resampled audio were that it would be fine for
audio coming from another room, but not for use in the same studio.
However it appears resampling can work just fine for most things provided
a single device is used for any one instrument (a drum set for example).
I am wondering if I have the ears/mics/sound reproduction equipment to
hear the difference. I will have to experiment once I get my studio set up
again.
Thank you for your answers.
--
Len Ovens
www.ovenwerks.net
One more... if a group of inputs that start
synced are resampled, do
they have a better post resample channel to channel sync that
accross two intefaces with different sample rates before resampling?
(this is sort of what the above questions are getting at though
there amy be other issues as well)
I don't understand the question... could you rephrase it ?
10:40 a.m.
New subject: questions about resampling - was - jack and the merging of soundcards
On Fri, Nov 21, 2014 at 08:11:29PM -0800, Len Ovens wrote:
So I was asking if the audio phase/sync would be
preserved if a
stereo pair was split across two devices, one out of sync with the
other and resampling being used to sync the two together. It appears
from your other answers, that there may be noticable artifacts.
It all depends on how big the errors are. The test results at the
end of my LAC 2012 paper provide some indication of this. From these,
if you compensate for the static delay difference, the remaining
variations are a few microseconds. Four usecs at 20 kHz means
around 30 degrees in phase, leading to the 0.3 dB loss in a mono
mix mentioned before. I don't think anyone will notice.
Ciao,
--
FA
A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)
4:05 p.m.
New subject: questions about resampling
On Sat, 22 Nov 2014, Fons Adriaensen wrote:
On Fri, Nov 21, 2014 at 08:11:29PM -0800, Len Ovens
wrote:
No I don't think anyone would notice that. I was thinking more about the
stereo imaging. I know at some frequencies the brain senses direction of
the sound by the phase difference from one ear to the other. The last time
I did speaker listening tests (10 years ago?), I found very few that gave
good imaging anyway (and they were not the most pricy either). Maybe
things have gotten better. Maybe usecs of difference from one ear to the
other are not noticable too.
--
Len Ovens
www.ovenwerks.net
So I was asking if the audio phase/sync would be
preserved if a
stereo pair was split across two devices, one out of sync with the
other and resampling being used to sync the two together. It appears
from your other answers, that there may be noticable artifacts.
It all depends on how big the errors are. The test results at the
end of my LAC 2012 paper provide some indication of this. From these,
if you compensate for the static delay difference, the remaining
variations are a few microseconds. Four usecs at 20 kHz means
around 30 degrees in phase, leading to the 0.3 dB loss in a mono
mix mentioned before. I don't think anyone will notice.
6:01 p.m.
New subject: questions about resampling
On Sat, Nov 22, 2014 at 07:01:23AM -0800, Len Ovens wrote:
No I don't think anyone would notice that. I was
thinking more about
the stereo imaging. I know at some frequencies the brain senses
direction of the sound by the phase difference from one ear to the
other. The last time I did speaker listening tests (10 years ago?),
I found very few that gave good imaging anyway (and they were not
the most pricy either). Maybe things have gotten better. Maybe usecs
of difference from one ear to the other are not noticable too.
One microsecond means 0.34 mm (1/75 inch) at the speed of sound,
or 0.12 degrees change of direction for a frontal sound source.
So you need 8 usecs to change the direction by one degree.
Maybe you could detect that if your head were clamped in a
vise. Test at your own risk :-)
Ciao,
--
FA
A world of exhaustive, reliable metadata would be an utopia.
It's also a pipe-dream, founded on self-delusion, nerd hubris
and hysterically inflated market opportunities. (Cory Doctorow)
21 Nov
21 Nov
5:09 p.m.
On Thu, Nov 20, 2014 at 5:12 PM, Ede Wolf <listac(a)nebelschwaden.de> wrote:
Thanks very much. I do know about zita-a2j any may end
up using it for
recording smtpe through onboard audio, at least as an experiment, but for
standard audio I would like to avoid resampling whenever possible. Also,
maybe pure superstition, that a2j attached soundcard always feels as just
being second best, not treated equal.
However, if I may re-ask, as it has not been quite clear to me: While alsa
seems to be capable of handling two cards that way, would this also work
for jack or would jack refuse to fail for this trick?
if the ALSA layer behaves, JACK has no idea what is going on. But this does
require that it manages to start all N cards within a 1 sample timeslice.
Theoretically, it should be possible to do this in microseconds.
4001
days inactive
4002
days old
linux-audio-user@lists.linuxaudio.org
14 comments
5 participants
participants (5)
-
Ede Wolf -
Fons Adriaensen -
Len Ovens -
Paul Davis -
Simon Wise