Re: [linux-audio-user] My MIDI latency is crap

On Tuesday 09 May 2006 12:23, Florian Paul Schmidt wrote:
> On Tue, 9 May 2006 12:43:29 +0200
>
> Werner Schweer <ws@email-addr-hidden> wrote:
> > Its not that easy. If you play a software synthesizer, the received midi
> > events can only be processed on the next JACK cycle. To reduce jitter,
> > all events in MusE are delayed by two JACK cycles. So the minimum midi
> > latency is two JACK cycles. Immediate processing of midi events whould
> > mean to align them to JACK cycle boundaries which is something you do not
> > want. A musician can easily deal with even huge latencies but not with
> > jitter.
>
> Reducing jitter and providing a constant latency is the softsynth's job,
> Not MusE's [well, MusE should of course introduce as little extra delay
> as possible to minimize latency and jitter]. MusE should simply send out
> the MIDI event as soon as it has received it [we do still talk about
> MIDI _through_ right now, no?].
>
> If MusE actually added another two JACK cycles, the resulting latency
> would be four JACK cycles as the softsynth would add another two JACK
> cycles itself.
>
> The only case where MusE would have to care about this would be when
> itself were a softsynth. Or maybe when playing recorded tracks of which
> one goes to a hw synth outside the computer and the other goes to a
> softsynth. In this case it would make sense to delay the MIDI stream
> that goes to the hw synth outside the computer by two JACK periods
> because that's the delay the softsynth will have, too, resulting in both
> being again played in sync.
>
> Maybe we have a complete misunderstanding here, though.

I was talking about live playing an MusE internal SW synthesizer. For an
external SW synti you use a direct connection to ALSA. A proper designed SW
synti should then do the same thing as MusE does resulting
in the same latency. The fact is that every SW synth has at least a two
cycle latency.
A problem with an (MusE-) external SW synti is that
it has to use an extra RT thread to collect midi events in realtime. This
RT thread muse have a higher priority than JACK. If not, all your midi events
will be aligned to JACK period time. I suspect this is the main reason for
horrible midi timings some people experienced. My expectation is that the
upcoming JACK midi implementation will cure all this problems by eleminating
that extra RT thread needed for midi.

Of course only the sequencer can provide a constant latency across several
instruments. An external synthesizer has to provide a as low as possible
latency. The sequencer should do a "latency compensation".

>
> > > Additionally, it seems MusE doesn't use ALSA's timestamps for received
> > > events but manually looks at the timer when it comes around to reading
> > > the events. This will somewhat degrade the accuracy of received events.
> >
> > yes, thats true, but as the receiving thread runs with RT priorities its
> > negligible in practice.
>
> Flo

/Werner
Received on Tue May 9 20:15:03 2006