Audio Input Latency on Push 3S with a Real-World Live Set

[Clint Goss]

Just loaded my first real-world live set into my Push 3S. Was dismayed that the latency between Audio 1/2 Input and Audio 1/2 Output is a whopping 116 msec. Hoping there is something basic I'm missing, because at this point running live input through my Push 3S in live performance settings will not be workable.

Reducing the buffer size from 64 to 32 reduced latency to 113 msec.

The latency reduced to 9 msec when I load a new (empty) project, with the audio set at 48kHz / 64 samples.

So it's the size of my set ... which I don't understand. I thought that if the processor gets behind, you get audio glitches. Never heard of an setup that simply slowed down when it got overloaded.

I've Frozen about 15% of my tracks because they used Spire, but want to keep the rest thawed so I can adjust them as part of performance.

Of course, I could route my live signal through some hardware FX and directly to my output mixer. But that would prevent me from live looping (I've had poor results trying to MIDI Sync loopers) and effectively turn my Push into a background track (or a stem track) player.

Any ideas appreciated ...

** Background

I've been live looping and performing with others for 3 years using:

* a Sylphyo wind controller,
* a laptop-based rig with Cantabile (a live DAW),
* several hardware sound generators (VL70-m, CraftSynth2, XpressO), and
* RME interfaces (Babyface Pro FS or UCX-II).

After a *lot* of work on latency reduction, I was running under 10msec latency. This was mostly accomplished by using Direct Monitoring in TotalMix for the dry sound (combined with software-generated FX). The difference between 10 msec and 15 msec latency is *very* obvious to a wind player - just do a fast scale run and it is immediately obvious in your fingers. The lower latency has a dramatic effect on playability for wind-synth players ...

I've moved to Live 12 and the Push 3S over the last 8 months to be able to perform live solo without the spectre of a laptop crash. I am routing all rendered audio (as well as mic for "real world" flutes) through the Push for FX and live looping.

I've just assembled the first 4 songs of a Live Set and the 116 msec latency I'm currently getting is completely disorienting and unusable.

** Measurements

I used Cantabile to record a mic tap impulse coming directly out of a mixer to an RME UCX-II alongside that same mixer output routed through Push 3S to that same UCX-II.

My Reference signal chain was:

* Mackie Mix5 Phones output =>
* RME UCX II Analog Input 7/8 => Cantabile recording

The Push 3S Test signal chain was:

* Mackie Mix5 Main output =>
* Push 3S Audio Input 1/2 (Line, Balanced) =>
* Push 3 v2.02 Firmware 1.3.3, 48kHz, Sample buffer size: 64 =>
* Push 3 Audio Out 1/2 =>
* RME UCX II Analog Input 3/4 => Cantabile recording

The difference between the two recorded impulses (measured in Reaper) was 0.116 sec = 116 msec.

Reducing the sample buffer size to 32 reduced latency to 113 msec. I think this 3 msec improvement is expected, since the buffer size setting would control both the input and output buffers.

The latency is reduced to 9 msec when I load a new (empty) project, with the audio set at 48kHz / 64 samples.

[[jur]]

Are you using latency heavy devices like Spectral Resonator, or some M4L ones maybe?

[Clint Goss]

... like Spectral Resonator ...

Right. Extremely helpful. How did you know?

So I'm just getting up to speed with Delay Compensation and those latency displays in the info bar when you mouse-over the title bar of a device (one Inst Rack alone is registering 58.7 ms).

Turning on ...
Ableton Live => Options => [X] Reduced Latency When Monitoring
... completely cures my issues with Live on my PC, but ??this option is not available on Push 3S (is that correct)??
(I'm looking at https://forum.ableton.com/viewtopic.php?t=247549)

I think one option for me is to:
* Split my input signal (WindSynth => HW rendering boxes => Input Mixer => HW FX) into two
* Send one input signal to my output mixer
* Send the other input signal to Push 3S for live looping
* Remove software FX on my input tracks in Live and rely exclusively on HW FX on my input path
* Turn off Input Monitoring on my input tracks in Live.

This will (I think):
* sync my low-latency input signal with the high-latency output of Live and,
* for live-looped recordings, delay my live-loops to match the high-latency Live tracks.
... and things should always be in synch ....
Does that sound correct?

(BTW, I can't imagine trying to get through this with Ableton Tech Support ...)

[Clint Goss]

My new split setup is working well ...

Recommendation: If you're feeding your live sound through a Push 3S (for FX and looping), the max latency of any track in your set may be an issue. If you don't need the FX of the Push 3 for your sound (i.e. you can use HW FX), then split your post-FX live sound into two paths: Direct to the output (i.e. to an output mixer) and into the Push 3. This lets you play with no additional latency on your live sound and also loop with the proper latency to match your Push 3 feed.

Details: I split a stereo unbalanced live-sound feed using two Y-splitter cables to the Push 3 analog inputs and to a stereo channel of my output mixer. The Push 3 output goes to another channel on the same output mixer. For solo gigs, I use a basic Mackie Mix5 or, for performances that include other musicians, I use an Alesis MultiMix8 Firewire. I compared the Y-splitter setup with a setup using a passive Radial JDI Stereo splitter box and the signal quality was the same (i.e. no ground loop with my simple Y-splitter). All unbalanced runs are less than 2 feet. The Radial had the downside of dropping the transformer-isolated feed by 22dB, so the Y-splitter is far simpler and preferable.