Group Admins

  • Profile picture of warusfel
  • Profile picture of T.


Public Group active 12 hours, 25 minutes ago

User group for Spat, panoramix, ToscA , and ADMix

mitigating clicks and dropouts

Author 3 Subscribed Users |
Profile photo of attackrelease

Hello all, I wanted to find out from others what software/hardware setups you’ve found to be the most stable in terms of smooth spatialized playback. I realize there are many elements that contribute to the entire performance of a particular pipeline, but am curious where folks are finding the limits are with their own systems.

I am running a late 2013 Mac Pro (3.5ghz 6-core Xeon) with a Motu 828x + Behringer Ultragain Pro-8 DA via Optical ADAT. I have a 16 channel speaker array in my room.

In MAX, i have the I/O Vector Size set to 512 and signal vector size set to 128.

I’ve found that keeping all sound generation completely contained with Max usually gives the cleanest results. Loading up wav files or running synthesis in patches and sending these into SPAT then out to the interface rarely results in any significant dropouts.

Recently however, i’ve been building sessions in Ableton because of the added flexibility for composition and sound design. I’ll stem out the mix in Ableton to a virtual loopback device (JACK or Rogue Amoeba’s Loopback), picking up the audio streams in Max, spatialize with Spat, then send out to the Motu. I’ve found as i begin adding instruments and effects in Ableton, the likelihood of very intense dropouts/clicks becomes much more common, even after optimizing buffer sizes.

I love the workflow of creating compositions in ableton and being able to modify effects on the fly during live performances along with the spatialization effects, but have felt that i need to be very reserved with what i place in my Ableton session before the dropouts really make things unusable.

Have you found that dedicating separate computers to these tasks improves performance? I’ve noticed others mention that they will connect their laptop to dedicated spatialization computer using a Dante network which seems like it could be a much more powerful setup. Let me know what you have found works best in your experience.

Thanks, Gabe.

  • This topic was modified 1 year, 5 months ago by Profile photo of attackrelease attackrelease.
June 23, 2017 at 22:59 #22729
Profile photo of tannerupthegrove

Hi Gabe,

I am running a 2013 Mac Pro with similar specs with no dropout issues. Sources are any of the following, including combinations with OSX aggregate audio device:

128 channel Dante input via RedNet PCIe
64 channel Dante Virtual Soundcard
32 Channel Yamaha TF Rack
SFPlay~ [in-patch]
64 Channel Soundflower routed from a DAW on the same computer

Again, all have been successful with no dropouts. Using external sources adds some latency.

On the output side, an aggregate RedNet PCIe and DVS afford playback on 139 loudspeakers without hiccups. I prefer to use the Mac Pro only for spatialization, sending 48kHz 24 bit audio via Dante as well as control data with OSC. I can imagine a patch generating audio as well as spatialization might cause hiccups, but maybe the local routing is the culprit? Soundflower is worth a shot too. Cheers.

June 24, 2017 at 22:19 #22736
Profile photo of T.


Basically there are 3 options:
– two separate computers (connected with Dante, MADI, or any other multichannel protocol)
– one single computer and virtual routing (e.g. Jack, Rogue Amoeba’s Loopback, etc.)
– one single computer and hardware routing, e.g., all RME audio interfaces offer the possibility to “loopback” outputs to inputs (with 1 sample delay), therefore allowing to route audio from one application to another.

At Ircam, we have used all three approaches successfully, although it is clear that the second one is the most tricky
and prone to audio dropouts. (the robustness of such approach seems to depend on many many parameters (CPU, OS version, number of channels, buffer size, use of aggregate, etc.))


(Just a remark also: in many use cases, it is not useful to have ‘signal vector size’ < ‘i/o vector size’. This increases the cpu load, but does not reduce the latency)

June 26, 2017 at 09:59 #22738

You must be logged in to reply to this topic.

Log in now