diagnostic tools to trace the reason xruns are happening?

Ralf Mardorf ralf.mardorf at alice-dsl.net
Sun Jun 5 22:44:37 UTC 2011

On Sun, 2011-06-05 at 23:09 +0200, Hartmut Noack wrote:
> Am 05.06.2011 20:05, schrieb Ralf Mardorf:
> > On Sun, 2011-06-05 at 19:39 +0200, Hartmut Noack wrote:
> >> Am 05.06.2011 11:00, schrieb Robert Klaar:
> >>> What happens if you set the frames/period settings higher? I have similar
> >>> setup and can't get it to run on any lower than 1024 frames(if I want to
> >>> avoid xruns), but then I've set the sample rate to 48000. I've never noticed
> >>> any problems with the latency, it's at 46 or something now but I can't hear
> >>> any difference between this and something lower, can you? .)
> >>
> >> Yes I can.
> >>
> >> I work with musicians and my experience is:
> >>
> >> lower than 5ms: nobody noticed that in comparison with 10ms
> >>
> >> 5-10ms: everything fine, no complaints
> >>
> >> around 16ms: all but some singers realise that something is going on,
> >> faces grow longer, some demand changes.
> >
> > Ok, so we do agree that the Haas-Effect has nothing to do with audible
> > delay for a groove, but I don't agree that lower than 5 seconds isn't
> > audible. It is, regarding to a groove, e.g. played by a MIDI keyboard,
> > controlling a soft-synth. It's annoying, but a constant delay is
> > something a musician is able to handle, anyway, I do agree that up to 11
> > ms are 'quasi' inaudible. Pipe organs might have a delay up to 30 or 50
> > ms, I dunno, but musicians are able to play such an instrument. Just
> > jitter is a no go, especially for this case ...
> >
> >>
> >> For very big projects on lesser machines I run Jack with 30 or more ms
> >> latency. It is OK for mixing/editing etc but I do not record any
> >> overdubs with such settings.
> >
> > ...
> >
> > What's the problem with overdubs? You listen to the delayed sound, but
> > for you it isn't delayed, you play in real-time without monitoring the
> > delayed recording and the latency compensation moves it to the correct
> > time.
> The problem is, that I record most of the overdubs using software, that 
> either generates or processes the signal.

Ok, for generated sound there is a latency issue. I just wondered,
because you wrote about vocalist. Regarding to issues for singers, the
monitoring should be without processing.

> > Hence latency has no impact, but jitter would have, because jitter
> > can't be auto-compensated.
> I use to record overdubs by playing instruments/voice so we would talk 
> about audio-jitter here.
> That would be:
> If I record the very same audio-signal from a Jack-output to a track in 
> Ardour several times it would be predictable, that the recordings start 
> to be/sound affected by such jitter-effects.
> How many generations do you estimate until such effects would become 
> audible?

First generation, not as timing issue, but as phasing. This phasing
won't cause a phasing effect, but less transparent sound.

> >> I remember I did some years ago but it
> >> really was not fun and the results where not as good as they could have
> >> been.
> >
> > I don't understand? What happened? Latency only will become an issue, if
> > it isn't a fix delay, or if you need to monitor the delayed output, e.g.
> > for soft synth,
> exactly that, see above.
> > apart from that latency has no influence.
> >
> >> Any recent Linux sould allow settings for 16ms or lower for normal load.
> >
> > It's related to the hardware and work flow. ElCheapo cards might be able
> > to keep such latency when Periods/Buffer are set ex 3, jitter than might
> > be another issue ;).
> Yes, it is known, that cheap chipsets almost never provide exactly the 
> samplerate they are set up with and that they can drift.

I didn't know that, but I do believe that this is true, because ...

> To be honest: I do not hear any effects of that.

... I do hear a bad sound quality for most sound cards and I guess it's
regarding to (audio , not midi ;) jitter and sync issues.

> Please correct me, if I am wrong: the soundcard is only the 
> player/recorder. Jack processes the signal internally whithout using or 
> even needing the sound card at all. Thus it is no great suprise, that 
> material, that I process while Jack runs with a HDA-chipset fails to 
> degenerate and sounds not better or worse if I play them the other day 
> on a envy24-chipset or a Presonus Firebox.
> I do hear the differences of a ADC/DAC for 30 Euros compared with a 
> converter for 30 Cents ;-)
> >
> >> If it does not, I try the standard-procedures to fix it, if this does
> >> not help I switch the distro.
> >
> > And first of all, don't care that much about latency, because latency
> > very seldom is an issue, just jitter could become an issue.
> I agree, that audible jitter would be an absolute show-stopper.
> Anyway I never heared such effects in any recordings I made in the last 
> 7-8 years. I only had audible effects in the form of xruns and in 
> test-scenarios with insane settings or with alpha-software or a severely 
> misconfigured system.

Yes, glitches are bad, but IMO (excepted for computer generated sound)
latency isn't important, so it should be possible to increase the
latency to get no glitches. Btw. I completely agree that when playing a
virtual synth, a latency around 10-11 msec is ok and ElCheapo cards
already enable such latencies. If not, something is broken.

Unfortunately hardware and Linux always cause issues. I bought a RME
HDSPe AIO and will return it to the dealer this week, because it's a
PITA to get the card working. At the moment I've got no idea what sound
card I should buy instead :(. I'm using Linux only, but good sound cards
seems to be supported for Mac only, resp. sometimes for Windows too. I'm
not fine with the sound quality of M-Audio and similar gear.

Ok, you're right regarding to latency, I misunderstood what you've
written. Anyway, as long as people avoid to play a soft synth by a MIDI
keyboard or they avoid to monitor processing, latency isn't an issue, at
least Ardour will compensate latency.



More information about the Ubuntu-Studio-users mailing list