Apogee Symphony Users... 1.6 ms of latency? Hype, or the real deal?

Echoing jnashguitar's sentiments:

1) Monitor through your audio interface as you're doing now. I know you said it's a PITA, but I can't help but think that a knowledgeable guy like you should be able to get down a workflow where it becomes second nature.

2) I second the purchase of an analog board (of some kind or other) for monitoring purposes.

Anything else is going to make you:
• spend more money
• have to deal with monitoring latency

As I see it, unless the convenience of doing everything ITB is worth the price of all that hardware you listed (plus the inevitable latency), I'd go the E-Z 'n' Cheeep route and do one of the above.

🙂 iS

"And move one of the speakers 2m away, and it will be obvious that there are two sound sources (i.e. your ears/brain will resolve the sounds individually)."

?

Not according to the Haas precedence effect. Two sounds need to be 50ms apart for them to be perceived individually.

"(I experienced this effect while fine-tuning the position of my stereo speakers in my living room. The goal, of course, is to form an isosceles triangle between the speakers and the listener. An identical sound originating from both speakers should reach the ears at precisely the same time at precisely the same volume. Any deviation from that weakens the "phantom" image between the speakers. My experience was that when those distances were more than a few inches off, the difference was audible.)"

Sure. Turn your head a fraction of an inch and the image shifts. Or delay delay it by .1 ms and you'll hear it shift; delay it by 1ms and it shifts all the way over to the earlier side.


"Also, consider that the ear/brain's ability to identify and localize sounds depends essentially on latency cues (typically called the "interaural time difference"). It has been well-researched that the ear can distinguish a time shift between the two ears when it is a mere 10 MICROseconds. Not exactly the same thing as feeling the latency of a DAW, but the point is that the ear is amazingly sensitive to time and phase, so unless you've actually tried a blind A/B test yourself, don't make the mistake of underestimating the ear!"

But that really doesn't have anything to do with latency through a DAW. The only issue is phasing with the live sound - or if you're playing V.I.s, the delay between playing and hearing the note.

*

To me the central question is whether the Symphony card's latency *through Logic* is any faster than anyone else's - not the "Logic integration" (which I find to be a very weak argument, considering that all it does is recall settings and give you a grey screen inside Logic that does most of what the color one you can access more easily by sticking in your dock does; what is impressive about Apogee is the really high quality of their interfaces). And I don't know the answer about the latency.

derivativemusic wrote:
"And move one of the speakers 2m away, and it will be obvious that there are two sound sources (i.e. your ears/brain will resolve the sounds individually)."

Not according to the Haas precedence effect. Two sounds need to be 50ms apart for them to be perceived individually.

Haas effect is a good point, and I wasn't clear enough. That type of sensory inhibition is, of course, real, and it gradually diminishes at distance increases. Personally, from experience, I find I don't need nearly 50ms of delay to create a "doubling" effect between the two speakers, but it really comes down to how you define "perceiving individually." At 50ms between the sources, I think we'd all agree that there are two sounds. At 1ms between the sources, it's essentially impossible to hear the delayed sound at all. In the range of 5ms, personally, the way I hear things is that the stereo image collapses entirely--it becomes obvious that there are two speakers (i.e. I hear sound coming from left and right, as opposed to from a "phantom" center). That's what I was trying to describe.

Good point to bring up the Haas effect, though. And if anything, I think it underscores the fact that there are complex psychoacoustic effects at work when sound comes from multiple sources with minimally different timing (e.g. live sound and latent monitored sound).

The only issue is phasing with the live sound - or if you're playing V.I.s, the delay between playing and hearing the note.

Exactly--this is what I and others are talking about. And there is always phasing with the live sound, and it's typically made much worse by any kind of digital latency. Isolating headphones can reduce the effect, but I've found that even 1ms of latency is easily distinguished from zero, especially on vocals.

To me the central question is whether the Symphony card's latency *through Logic* is any faster than anyone else's - not the "Logic integration" (which I find to be a very weak argument, considering that all it does is recall settings and give you a grey screen inside Logic that does most of what the color one you can access more easily by sticking in your dock does; what is impressive about Apogee is the really high quality of their interfaces). And I don't know the answer about the latency.

Yeah, I don't know that either. The Apogee tech quoted earlier was saying Symphony would give 4ms of latency at 44.1. If that's accurate, then it's not really any better than MOTU Firewire, at a given sample rate. Maybe the point is that at 96k, Symphony would give under 2ms, while other interfaces can't keep up with that small a buffer ? Again, I'm only guessing--don't know the actual performance. Pretty sure my MOTU (Firewire) can't do a 32 sample buffer at 96k, though, so if Apogee can, that might very well be lower latency. Be nice to be able to get that low latency at 44.1, though...

James

Jim Frazier wrote:
Neither of you is suggesting I even bother trying to monitor through Logic, and I certainly understand WHY you're suggesting that.

Actually, though, as much as I dislike latency, I end up monitoring through Logic much of the time. It's just so dam [sic] easy, when I want to get to work as quickly as possible. When I'm tracking something critical, I'll always set up the analog mixer, but the 4ms of latency I get through Logic is totally usable. I find it sounds a little odd when compared to analog, but after a few minutes of tracking, it's easy to ignore 4ms of latency.

I know it sounds contradictory. On the one hand, I'm asserting that 4ms of latency causes serious phase artifacts in the headphones, and that it disrupts timing, while on the other hand, I'm saying it's easy to ignore. I guess that's the dichotomy between musician and engineer. When I'm being a picky engineer, 4ms of latency seems like a huge PITA. When I'm focusing on the music, I can ignore things much more serious than 4ms of latency. Like when I'm so focused on what I'm playing that I completely miss naked women dancing in the audience (sadly, this has happened more times than I can count... 🙂

Thanks for your input guys... I got to do some thinking here...

Good luck! Best by far would be an integrated analog mixer in the audio interface, with the ability to control that mix via software. Zero latency and simple to use. Too bad, but I don't think anyone makes that. Anyone want to start a hardware company ?

James

The Apogee tech quoted earlier was saying Symphony would give 4ms of latency at 44.1. If that's accurate, then it's not really any better than MOTU Firewire, at a given sample rate. Maybe the point is that at 96k, Symphony would give under 2ms, while other interfaces can't keep up with that small a buffer ? Again, I'm only guessing--don't know the actual performance. Pretty sure my MOTU (Firewire) can't do a 32 sample buffer at 96k, though, so if Apogee can, that might very well be lower latency. Be nice to be able to get that low latency at 44.1, though...

This was the point that I made earlier on. Jim's post included the question, "is it hype"? The answer is, IMO, "yes, it's hype to the degree that you ONLY get the low latency at 96K. At more nominal sampling rates (i.e., "more routinely used") of 44.1/48, the latency drastically increases to the point where, as you pointed out James, it's no more competitive than other gear.

Figure that doubling the sampling rate will halve the latency. So 1.6ms at 96K translates to 3.8ms at 48K. And at 44.1K it would be slightly more than that.

ok guys,
here is how this all goes.
on a very basic system with no latency introducing plugins
Latency = (IO_Buffer/Sample rate)*2
this is why when you increase the sample rate you decrease the latency, however, of course it means that you use more CPU power !!
now this is just pure digital latency of your system, but in real life you have to add to it the latency of the actual AD, DA convertors cause as mentioned before not all converters are created equally.
The reason why we multiply the above formula by 2 is to account for both Input buffers and output buffers.
This formula is universal and applies to all sound cards.
now when drivers are poorly written, they can add more latency to this basic amount, and if there can be even more latency added by poor hardware design, although most current sound cards don't add latency because of hardware.

That's the first part of the formula. There's more to add, though:

- any plug-ins;

- the converters themselves;

- and any latency introduced by the interface and its drivers.

It's the last one that's especially interesting (to repeat): is the Apogee Symphony's latency better than any other interface? Part of the answer may depend on the efficiency of its driver (assuming it's not being used as an aggregate device), and possibly how much work the card itself does in order to let you lower the buffer without straining the computer.

But in all honesty I'm less convinced of the last point than the conventional wisdom would lead one to believe. In my experience pretty much all FW devices have the same latency, and I'm not sure that in CoreAudio there's a lot of difference. I'm open to being convinced otherwise, however.

I should add that there's definitely a serious difference between drivers and cards on Windows machines. But they're using ASIO rather than CoreAudio, so all bets are off.

Interesting discussion, one thing I'd like to add.

You will definitely have small phase/timing issues between the acoustic sound of an instrument and headphones fed from a digital system that introduces latency.

But with the possible exception of voice, even on an analog monitoring system, you're still going to have timing issues anyway. No latency will only match up if the instrument is right next to your head.

The speed of sound in air is 1129 feet per second, so every foot of distance adds about a millisecond.

If you hit a snare that is three feet away from your ears but half a foot from the mic, with no latency, the headphone will be 2.5 ms earlier than the acoustic sound. For a kick that's 5 feet away, 4.5 ms early.

Unless you get lucky or actually tune a delay to the distance, there's virtually always going to be a timing and phase difference between the acoustic sounds and headphones.

Obviously the shorter the latency the better, but I'm skeptical that a 1.6 ms delay is that noticeable in most situations. It's the difference between a guitarist or bass player having his amp 5 feet from his head versus 6.5 feet from his head. Now if only they could get the latency that short at 44.1.

iSchwartz wrote:
This was the point that I made earlier on. Jim's post included the question, "is it hype"? The answer is, IMO, "yes, it's hype to the degree that you ONLY get the low latency at 96K. At more nominal sampling rates (i.e., "more routinely used") of 44.1/48, the latency drastically increases to the point where, as you pointed out James, it's no more competitive than other gear.

Yeah, and it's weird enough to make you wonder why, eh ? Are we certain that Symphony gets its lowest latency at the highest sample rates ? I don't know the inner workings of their drivers or of CoreAudio, but that doesn't make a lot of sense to me.

Important to remember that a buffer setting of 32 samples at 96k equates to the same *period of time* as a buffer setting of 16 samples at 48k. In both cases, you're constraining the OS to perform a job in X amount of time, and at 96k, that job is twice as large. Stands to reason that if audio hardware can run a given buffer setting at 96k, it should be able to run half that buffer size at half the sampling rate, and the result should be somewhat less load on the CPU.

So if Symphony gets 1.6ms latency at 96k, I wonder why it doesn't offer half the buffer setting to allow the same latency at 48k ? Not immediately obvious to me why that should be difficult. Maybe our information is wrong ?

James

I was thinking the same thing. If a hardware setup can run at a 32 sample buffer at 96k, shouldn't it have an even easier time running at 16 samples/44.1 or 48?

The process is going twice as fast at 96 than it is at 48. That's something we can take at face value - it's the same for every converter.

But it doesn't matter to me personally, because as a friend of mine puts it: 96k can kiss my as s. 🙂

I'm not interested in getting half the performance out of each computer, thank you very much.

davidmesiha wrote:
Latency = (IO_Buffer/Sample rate)*2

Using your formula, the result is an exceedingly low value. For example:

buffer size = 32 @ 96K sampling rate:

(32/96000)*2 = 0.000666666666666

... a latency of a litte more than 6/10,000ths of a second? Something doesn't seem right, because a single sample at 96K lasts for much longer than that -- 0.01041666667 seconds (1000/96000). David, can you please expand upon how to use this equation to calculate latency?

Message was edited by: iSchwartz

iSchwartz wrote:

davidmesiha wrote:
Latency = (IO_Buffer/Sample rate)*2

Using your formula, the result is an exceedingly low value. For example:

buffer size = 32 @ 96K sampling rate:

(32/96000)*2 = 0.000666666666666

... a latency of a litte more than 6/10,000ths of a second? Something doesn't seem right, because a single sample at 96K lasts for much longer than that -- 0.01041666667 seconds (1000/96000). David, can you please expand upon how to use this equation to calculate latency?

I think the formula is right, iS... just a unit problem:

In/Out 32 sample buffer at 96k is (32/96000)*2 = 0.00067 seconds = 0.67 milliseconds

A single sample at 96k is 1/96000 = 0.0000104 seconds = 0.0104 milliseconds

Yeah, 0.67ms is very small, and that's an important point. With a 32 sample buffer at 96k, the software buffering starts to become less relevant... most of the total latency will be due to the A/D/A conversion. If interfaces can start running with 16 sample buffers at 44.1k (which seems doable, given current CPUs), the latency playing field will be fairly even between host-based and dedicated-hardware-based DAWs. To get the latency much lower, the digital conversion will need to be sped up.

James