Measuring latency

[midnightsun]

Okay we all talk about latency of our systems for better or worse. As I have researched this issue I realize that there are a number of different approaches to determining "latency." For those of you have bothered to measure the latency of your systems I would like to hear exactly how you did your measurements.

[matt thomas]

play a signal out the outputs, plug the outputs into the inputs and record the signal, see how far apart they are.

narco

[Pr.tiouz]

I made the test last night in logic audio 7.1.1 OSX 10.4.5 bipro G5 2.5, metrichalo 2882 latest software package.

well i'm very impressed i thought i would have to compensate for converters latency not taken into the delay compensation of logic but it's dead on absolutely not a sample of delay the recorded track is perfectly matching the played file. same with the metronome recorded, it's right on the bar.
so i decided to went further to know if i can use plug-ins while i'm still overdubbing, so i inserted all kind of plug-ins, native, powercore, on tracks on bus on output using sends and so on, and it was still dead on (i thought before that i couldn't do that), logic was set to full PDC (the one including bus), and it was perfect.
i'm now wondering if it's the same on 7.2.2, i will make the jump in febrary.

Matthieu

[Pr.tiouz]

hoops sorry i forgot, it was one sample ahead with the TC MD3 on the output

tutt

[Pr.tiouz]

I realize i'm totally out of topic, you were talking about the latency from input to output, of the DAW not the one from output to input wich Narco was refering to.

i plan to measure it precisely these days when i get the time, i will use spectrfoo which has something called transfer function wich include a tool to mesaure the time delay between two signals. i this way i will know the real latency i have on my system and not the one you find by converting buffer size to milliseconds.
I will report back in a few days.

regards
Matthieu

[theother]

Originally Posted by narco

play a signal out the outputs, plug the outputs into the inputs and record the signal, see how far apart they are.

narco

wrong!

That's what I did first and it does not measure latency which is easy to prove when you change the buffer size and notice no difference when you record the signal again.

What you get though is the latency for analog inserts, in case you use them with your digital built in mixer.

The proper way to do it is (what would be the latency what you hear in your headphone during recording):

Feed a signal on Input L (I just clap my hand on an open mic line)
Record Input L onto Audio Track1 and monitor your recording from audio Track1 on Output R.
Patch Output R to Input R (analog) and record Input R simultaneous to Audio Track 2 with Track1.

Compare both Audio Tracks = latency.

[theblue1]

Originally Posted by narco

play a signal out the outputs, plug the outputs into the inputs and record the signal, see how far apart they are.

narco

Well -- that tests what some call "track misalignment" -- uncorrected latency, if you will, with regard to overdubs vis a vis previously recorded tracks.

And that is, of course, a very critical measurement. My own hardware and drivers have a 356 (or is it 355, now, I forget) sample misalignment when used both in Sonar (5) and in Mackie Tracktion (1) -- but Tracktion offers a kind of "ping loopback" test and auto-calibration utility that, once set, will compensate for that particular misalignment.

But there are a number of latencies typically involved in digital audio. There's the latency involved in both AD and DA conversion (primarily related to filling appropriately sized i/o buffers)... there's the latency involved in processing the audio in the computer... there's the latency involved in auto-plug-delay comp. Some of these "overlap" and some are additive.


The main places where monitoring latency comes to the fore, of course, are where you're monitoring "through-the-box"... say with virtual instruments -- or perhaps input monitoring with FX, say, to use an amp sim (which has a cumulative latency from AD and buffering, processing in the box, and then buffering and DA).

Anyhow... there are a lot of latencies all over the place. Some don't matter if you're monitoring through a zero latency device (something with an analog mixer) or perhaps a "near-zero-latency" digital mixer or NZL monitor output on a converter (although one man's "near-zero-latency" can be another man's "it just don't feel right!").


But that track misalignment issue (mentioned at the top of my post) is another matter. In my highly informal (and very spotty) ongoing 'survey' of other users' experiences, most of the people who've done ping-loopback tests have been surprised to find that their O/Ds are misaligned by anything from a sample or two to 20 or 30 ms or more, with most being in the "grey area" beneath 10 ms. (For me, my 8 ms misalignment is too much for critical parts like percussion... even though that's the time it takes sound to travel about 7 and a half feet or so... it often feels "out" to me -- particularly with percussion but also even with some guitar comping. [I AM a sloppy player, I think, but I want it to be MY slop... not the arbitrary slop of device misalignment.)

[theblue1]

Originally Posted by theother

wrong!

That's what I did first and it does not measure latency which is easy to prove when you change the buffer size and notice no difference when you record the signal again.

What you get though is the latency for analog inserts, in case you use them with your digital built in mixer.

The proper way to do it is (what would be the latency what you hear in your headphone during recording):

Feed a signal on Input L (I just clap my hand on an open mic line)
Record Input L onto Audio Track1 and monitor your recording from audio Track1 on Output R.
Patch Output R to Input R (analog) and record Input R simultaneous to Audio Track 2 with Track1.

Compare both Audio Tracks = latency.

Right on the latter.

One point on the former, though: you're right as far as you go -- but the analog outputs are the "Now" you are hearing when you're overdubbing. So anything you overdub to THAT particular timing will be misaligned by whatever that amount is. IOW, if your ping loopback test (previously recorded track recorded onto a new track via analog out and compared) shows x ms misalignment -- all your overdubs will reflect that misalignment, by the nature of the process.

Worse, while most folks will probably lay down a rhythm guide first (drum or click) and try to key off that... say you overdub a bass as your second track (on my rig that would provide an 8 ms misalignment unless corrected)... NOW what if, when you overdub your guitar part, you take your rhythmic cues from the bass? It's 8 ms "behind" the drums. Your guitar part will be 8 ms "behind" the bass. Suppose your singer listens to the guitar part for his/her rhymic cues (instead of the bass or drums)... the singer is now 24 ms "behind" the drums. This is, of course, a somewhat unlikely worst case scenario -- but you can see how it adds to rhythmic imprecision unless corrected.

Because so few people pay attention to this -- or are even aware of the issue (a lot of people can't seem to wrap their head around the relatively simple logic of it, God love 'em) -- I can't help but feel that a lot of folks who complain about digital audio not sounding "right" may be unconsciously picking up on this often subtle problem.

BTW... I THINK Cubase's "hardware compensation" with its ping-loopback calibration test works similarly to Tracktion's... but I'm not certain. (As I understand it, the Cubase utility is purposed toward compensating for outboard digital FX box latency -- but it seems likely it could be used to adjust for any "straight" track misalignment by the same token.) And, supposedly, Sonar has added some form of track misalignment compensation in Sonar 6 (though I'm still using Sonar 5).

[matt thomas]

Originally Posted by theblue1

Well -- that tests what some call "track misalignment" -- uncorrected latency, if you will, with regard to overdubs vis a vis previously recorded tracks.

And that is, of course, a very critical measurement.

and thats what I was refering to (obviously)

narco

[theblue1]

Originally Posted by narco

and thats what I was refering to (obviously)

narco

Yah... unadjusted track misalignment is one of my personal pet peeves. When I had a PCI based card and my misalignment was under 5 ms I was philosophical.

By the time I got my MOTU 828mkII (a FW box, ov course) I sort of hoped all that stuff was a thing of the past and that things would line up in the DAW.

Wrong again...

I'd tracked a bongo part and -- while I'm certainly not the most precise hand percussionist around -- when I played back what I thought was mostly a keeper I was amazed... the whole thing just sounded off... every single hit.

I decided right then and there to test my overdub alignment per (in essence) a ping loopback and was mildly shocked to see around 8 ms of misalignment from my analog copy to the original.

I was also kind of amazed that 8 silly little milliseconds could make enough of a diff to sink the whole track... I mean... it's not unlike sitting the bongo guy about 7-1/2 feet from the mic, for heck's sake. But when I nudged my bongo track the measured 356 sample difference it sounded "right" (I'm STILL not a great bongo player -- but at least it sounded like what I put down! )


Anyhow, I'm hoping CW really did make good on Ron Kuper's promise to address the issue (with at least a Tracktion-like calibration utility). According to a couple early reports they did. But this ish is NOT at the top of everyone's list...

[midnightsun]

Very thought provoking responses. For me what really matters is the misalignment which can be additive as others have pointed out. That means that for overdubs it should be lined up as if all of the parts were played simultaneously in the same "groove" so to speak.

A variable that no one has mentioned is the actual latency of the musician. We are dealing with very small numbers in the hardware of our studios. The human ear has mechanical component (transducer) in front of both peripheral nerves and brain.

The other day someone recommended a method of calculating "practical latency," which given this discussion is really more of a measure of practical track misalignment--
Lay down your own click track with a couple of drum sticks on track one. Then while monitoring track one, play right on the beat and record to track two. Compare the difference between track and one and two. This differrence becomes the number that is necessary for adjustment to put the new track in the "groove." This approach made sense to me from a pragmatic standpoint.

[5down1up]

converter latency depends on the converter itself and the samplerate.

means every signal you record will be late. i dont compensate for that, because EVERY played signal will have that delay amount in fact of A/D conversion.

measuring plug in delay, routing delays etc. insert the desired plugin, bounce it down to disk, compare the original + the bounced file. that way you can measure the delay the plugin is creating. same with busses etc.

i keep the buffer size mostly the same, sometimes when mixing i increase it to max.

PT mix user here running PT & Logic

Hope that helps ?

[theblue1]

Originally Posted by midnightsun

Very thought provoking responses. For me what really matters is the misalignment which can be additive as others have pointed out. That means that for overdubs it should be lined up as if all of the parts were played simultaneously in the same "groove" so to speak.

A variable that no one has mentioned is the actual latency of the musician. We are dealing with very small numbers in the hardware of our studios. The human ear has mechanical component (transducer) in front of both peripheral nerves and brain.

The other day someone recommended a method of calculating "practical latency," which given this discussion is really more of a measure of practical track misalignment--
Lay down your own click track with a couple of drum sticks on track one. Then while monitoring track one, play right on the beat and record to track two. Compare the difference between track and one and two. This differrence becomes the number that is necessary for adjustment to put the new track in the "groove." This approach made sense to me from a pragmatic standpoint.

Yeah... my track misalignment is always about 8 ms... my OWN rhythmic misalignment is WAY all over the map.

I can tell you that without testing.

Just the same, I don't need any MORE misalignment from my DAW, if I can help it, so I nudge my overdubs the appropriate amount.

As you suggest, the ideal -- I wish it was a baseline -- in recording overdubs is for them to align precisely with what is coming out the analog output of your converter at the same moment ('cause that is "now" for the person doing the overdub).


I don't usualy whine about the "good old days" but that's one thing that pretty much didn't get out of whack on a single tape machine -- alignment from one track to the next (because the simulsync record head doubled as a monitor head for record mode).

I mean... before dealing with the disconnect between DAW manufacturers and AD/DA converter manufacturers (and consequently driver writers) I would have NEVER DREAMED that people making supposedly professional grade gear/software would shrug and say, "What's a few milliseconds? Besides it's their fault!"

But that's pretty much what they do. Or have done in the not so recent past.

[theother]

Originally Posted by theblue1

Yeah... my track misalignment is always about 8 ms... my OWN rhythmic misalignment is WAY all over the map.

You are right, this can turn into a serious problem and I never thought about it. I always thought Cubase/Nuendo would correct the delay automatically.

I have to investigate this further!

[theother]

OK, I checked this track misalignment thing out.

Cubase 4 DOES compensate for the internal latency when recording.

What is doesn't do is compensating for the delay of the converters. (How should it know...)

So you have to measure the delay (re-recording an audio track) and then enter it manually under DEVICE SETUP>VST AUDIO SYSTEM>RECORD SHIFT (in samples)

Voila!! Everything is excatly in time now!! Wonderful!!

Thanks for telling me theblue1, I would never noticed that all my recording would 'fall back' in time gradually otherwise!

[theblue1]

And thanks for confirming that Cubase's hardware delay comp can be used to adjust for it!

For many, it may be a small enough amount that they decide to ignore it.

As at least one company rep has seemed to explain the lack of 'automatic' converter latency in some systems, the driver should supply accurate latency info to the host app. And then the host app should be able to line everything up.

But in anecdotal user test after user test, people have tended to find misalignment. There are no doubt exceptions -- at least if this "system" ever works correctly.

When some of us brought this up last year in a MASSIVE thread over at Harmony Central, Ron Kuper, CW's main man actually came in to argue what seems to be the DAW manufacturer's basic position: this mismatch comes from converter latency; if the drivers don't report accurate latency information, how can the DAW do anything about it?

But a number of us pointed out that -- one way or another -- all these people (including even Kuper himself with a a tiny but still there 1 sample "misalignment" on O/D's compared to previously recorded tracks) are having a situation where their overdubs don't line up.

At some point it doesn't matter WHOSE fault it is... if the job isn't getting done properly -- and that's over a wide range of hardware and installations -- then it seemingly has to fall to the software makers to provide us the means for compensating -- just as most of them have finally given us plug delay comp.


Mackie's implementation seems to work very well. It's a matter of pushing a couple buttons and it measures the misalignment and sets up the comp for you. I'd like to think that the fix in Sonar 6 that Kuper promised is at least as good as that. Sounds like Cubase's comp facility is similar. (Tracktion also gives you the option of entering a different comp value, though it does insert a default value.)



But... yeah... next time you read a post about someone whose overdubs/tracks just don't sound right to them but they can't figure out just what it is... you never know... it just might be this issue.