Highest latency before you start to notice

[The_BPP]

I did a bit of an experiment, last night, on a cheap Windows tablet...

Whilst playing a piano VSTi, I didn't really start to notice latency until it was around 15ms (in total) or higher.

I wondered if I was less or more sensitive than others?

I remember reading an article with Trevor Horn saying some of the session musicians he used noticed a 10ms delay. He clearly seemed to think this was an unusually small delay to perceive.

[Pete Kaine]

When discussing it with clients we tend to say drummers need around 10 μs, then guitarists and keys around 12 μs and vocalists don't really notice to around 15 μs.

Of course everyones different, and I imagine if you had to run a room full of session players that they'd all prefer to be working around 10 μs or below to ensure nobody has any complaints.

[Wonks]

Do you really mean Microseconds, Pete?

[Wonks]

Beware simplistic numbers. You've got the input processing latency, a very small processing latency within the DAW for your software instrument, and then the outgoing processing latency (which you don't normally get given numbers for), which is often close to the input latency value.

Depending on the AI used, the latency values reported bu the interface software may or not be particularly accurate. So reported small latencies may really be larger in real life.

And don't forget that every foot away from the speakers you are, you are going to add another 1ms of delay.

Look back 15 years or so and SoS articles were reporting that 12-13 ms latencies weren't noticeable. With the right AI and a decent processor we can now get latencies in the 2-3ms region.

[Dave B]

Apparently, according to Trevor Horn, the late, great Chris Squire (bassist in Yes) could hear timing discrepancies down to 9ms. This was discovered when they scoped the Synclavier sequencer and realised that it could be up to 40ms off the beat as NED didn't think it was that important.

[Pete Kaine]

Sorry, was stright after a DPC thread elsewhere, I meant ms.

[dmills]

The biggie tends not to be absolute latency so much as latency jitter, as long as the time between pressing the key and noise happening is substantially constant you can get away with much, but the moment it starts to differ note to note, it all goes very wrong even at the single digit ms level.

Early digital synths tended to run with an ungodly combination of large buffers and a philosophy of processing everything at the START of the next buffer (Thereby minimising AVERAGE latency as far as possible), so latency could vary over the full length of a buffer.

The more modern approach is to time stamp events with where they occur in the currently playing buffer, then place them at the same time in the NEXT buffer, this gets you constant latency rather then lowest possible average, which is far less of a problem.

Putting numbers on these things is **HARD** as it really needs to be tested double blind, confirmation bias is huge in this sort of research.

Regards, Dan.

[John Willett]

With digital radiomics - they used to say it should be 2.5ms or less.

Only one I know of has this figure; which is the new Audio Limited 1010 system at 2ms.

[The_BPP]

Apparently, according to Trevor Horn, the late, great Chris Squire (bassist in Yes) could hear timing discrepancies down to 9ms. This was discovered when they scoped the Synclavier sequencer and realised that it could be up to 40ms off the beat as NED didn't think it was that important.

That must have been the quote I was thinking of... I was nearly right I always wondered how they measured it.

I bought a Windows 10 tablet with an Atom processor, and wondered if it would be fast enough to use live as a VSTi player... So, I installed Kontakt 5, loaded up a couple of pianos, and got the latency down to 14ms without any clicks or pops.

This was perfectly playable, and felt no different from playing a standard stage piano, so that's a result! I've no wish to bring expensive gear to some of the shïtholes I play in.

[Martin Walker]

The biggie tends not to be absolute latency so much as latency jitter, as long as the time between pressing the key and noise happening is substantially constant you can get away with much, but the moment it starts to differ note to note, it all goes very wrong even at the single digit ms level.

Spot on Dan - I remember investigating latency jitter a long time ago in an SOS 2-parter September/October 2002:

http://www.soundonsound.com/sos/sep02/a ... an0902.asp

http://www.soundonsound.com/sos/oct02/a ... an1002.asp

Martin

[Mixedup]

Pete's right in that it varies from instrument to instrument. Despite my best efforts, I've never been able to set the latency low enough to be happy playing BFD live from my V-drums, for example, whereas playing guitar through an amp sim doesn't seem problematic at all (at least in this respect... the sound and responsiveness of amp modellers compared with amps is a different topic entirely!). For vocals... well, there needn't ever be any latency there, as I monitor what's going in and don't use plug-ins for any comfort verb etc. And piano? Well, we all know that keyboards are Satan's work, so I couldn't possibly comment.

[The_BPP]

Pete's right in that it varies from instrument to instrument. Despite my best efforts, I've never been able to set the latency low enough to be happy playing BFD live from my V-drums, for example, whereas playing guitar through an amp sim doesn't seem problematic at all (at least in this respect... the sound and responsiveness of amp modellers compared with amps is a different topic entirely!). For vocals... well, there needn't ever be any latency there, as I monitor what's going in and don't use plug-ins for any comfort verb etc. And piano? Well, we all know that keyboards are Satan's work, so I couldn't possibly comment.

A Singer = Someone who wants to get laid
A Guitarist = Someone who wants to pose
A Pianist = Someone who's in it for the music

[Mixedup]

A Pianist =

...someone who, like a guitarist, is happy with the hammer on but, unlike a guitarist, has yet to discover the joy of pulling off.

[Pete Kaine]

unlike a guitarist, has yet to discover the joy of pulling off.

Far too many jokes there to even know where to start.

[Mixedup]

unlike a guitarist, has yet to discover the joy of pulling off.

Far too many jokes there to even know where to start.

[The_BPP]

Believe me, after hours on our own practising, most pianists are very-much experienced in the joys of "pulling-it-off".

[Richard Graham]

Pete's right in that it varies from instrument to instrument. Despite my best efforts, I've never been able to set the latency low enough to be happy playing BFD live from my V-drums

Like you, I can deal with much higher latencies for guitar and keyboards than for drums. I think it's something about the physical connection with drumming, it's very immediate, and if the sound doesn't come out at the same time as I feel the impact in my hands and wrists, it feels like I'm drumming underwater.

I finally got the latency I wanted for drumming (i.e. unnoticeable) when I bought a 2012 MacBook Pro Retina i7. I'm running Kontakt (Abbey Road 60s and Studio Drummer) hosted in Reaper. I've stuck a Zoom TAC-2 Thunderbolt interface on it too, but to be honest, the built-in headphone out was pretty spiffing too.

I'm using the USB Midi out on my Yamaha DTXtreme IIs.

Could your MIDI interface be adding latency?

[Guest]

For me with guitar I start to notice at around 10ms, my brain hears it as a compressed sound for some reason. Then as the latency increases I hear it as a delay.

With keyboards I don't notice higher latencies so much, my brain seems to compensate quite well.

[Richard Graham]

A Singer = Someone who wants to get laid
A Guitarist = Someone who wants to pose
A Pianist = Someone who's in it for the music

A drummer: someone who hangs around with musicians.

[molecular]

How are people calculating this? Just out of curiosity...?

I was thinking about this last night, tracking an electric piano (audio, not MIDI), but listening on cans back out of the DAW.

I had to have the buffer at 64 samples before it became playable, added to the 10ms latency setting on the sound card making around 11.5 at 44.1khz. Although I felt I could still tell something wasn't quite right. I wonder if that calculation is accurate, though?

[Mixedup]

How are people calculating this?

I'm not, really. I mean I'll 'ping' to measure round trip latency for processing, but I've not really measured MIDI input device>audio playback timing. But having tried to get as low as possible (32 samples, 96kHz) drums are still off putting. But perhaps it's not just about the latency. There are so many things going on when you're trying to recreate a real experience (timing, physical position, sound, dynamic response... ) I imagine these get rather 'lumped together' in our minds when we perceive things as right or wrong.

[The_BPP]

How are people calculating this? Just out of curiosity...?

In Kontakt, it calculates the total latency in options/audio. I'm hoping this is accurate.

[Richard Graham]

How are people calculating this?

I'm not, really. I mean I'll 'ping' to measure round trip latency for processing, but I've not really measured MIDI input device>audio playback timing. But having tried to get as low as possible (32 samples, 96kHz) drums are still off putting. But perhaps it's not just about the latency. There are so many things going on when you're trying to recreate a real experience (timing, physical position, sound, dynamic response... ) I imagine these get rather 'lumped together' in our minds when we perceive things as right or wrong.

The latency was always the off-putting thing for me. Now it's down to a bare minimum the drums are a pleasure to play.

Having said that, it's also important to get the dynamic range right - I've done this mostly by setting the velocity curve in Session Drummer to "sensitive", and tweaking my triggers' velocities till they feel right.

You'll know if it's latency that's off-putting, if you don't find you have this problem using the kit's internal voices.

[molecular]

There are so many things going on when you're trying to recreate a real experience (timing, physical position, sound, dynamic response... ) I imagine these get rather 'lumped together' in our minds when we perceive things as right or wrong.

I'm sure that's true, and I think it must also depend on what you're actually trying to do with the instrument.

If there is a keyboard part that I know, and I am trying to record it and nail a solid take, I think I could have done that without worrying about it at what (according to the calculations above) would be between 11 and 15 ms.

What I was trying to do the other night was to write / improvise as part of the recording process, and so was much more heavily reliant on the feedback I was getting from the instrument and how that affected my relationship with it. The kind of concentration that makes recording a known part with a bit of latency possible, would only have been even more off-putting...

With regards to drumming, I used to tour with a mixed acoustic/electric drum setup to an in-ear click and I used to notice all manner of bizarre "psychosomatic" things with tempo and timing. The music was generally all set tempos for the duration of a track, but depending on when I changed hi-hat patterns in particular, or the nature of a fill, I would have to "slow down" substantially at specific points, or deliberately make hits very late in order to catch elements of the playback on the nose... I do think that when "in the zone", the mind does start to focus in eventually on a strange degree of timing detail that would be better left to intuition, and that's one of the things that makes using click tracks rather than ensemble playing a PITA from the musician's perspective.

A passing thought regarding guitars would be that maybe it is a bit less important to have super short latency because you get quite a lot of feedback from the physical interactions with the strings? With a V-Drum or a controller keyboard you get much less back through the fingers, so you are leaning more heavily on the sound coming out of the DAW for information about what you're doing...

[Sam Inglis]

In Kontakt, it calculates the total latency in options/audio. I'm hoping this is accurate.

Unfortunately, depending on what audio interface you're using, quite possibly not. The interface is supposed to declare its latency to the audio software, because it needs to know this in order to place recorded audio accurately on the timeline. However, many interfaces sneakily under-declare their true latency.

On Windows, the easiest way to measure the actual round-trip latency is using Oblique Audio's RTL Utility. On Mac OS there's no equivalent that I know of, but you can measure it by hooking up a loopback cable and re-recording a suitable audio signal (something with obvious transients) from one track of your DAW onto another. If they line up perfectly, then the reported latency is accurate. If the re-recorded signal is late then you need to add that delay to the reported latency to get the true latency.