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

Let's take your first point for granted. Natural delay between mic and amp, etc. So there I agree that there's a natural latency. Wow, we agree!

Mike Connelly wrote:
I guess what I don't get is the objection to a latency delay that's shorter than the delay caused by the acoustic sound travelling through the air - why someone who hears a three MS delay of the sound traveling through the air and is OK with it has a harder time playing when the sound travels through 1.6 ms of latency - when they're actually hearing the sound earlier.

But how is it possible to hear the sound earlier when the A/D and DAW-induced latency will inevitably make it later?

To put labels on this stuff for sake of brevity (imagine that -- me looking to be brief! LOL) the distance between whatever's being mic'd is "natural latency" that can never be compensated for other than applying one's skill at playing in time with the track. The add'l monitoring latency induced by digital system is "unnatural latency". I think the idea is to never introduce any unnatural latency into the picture.

I don't see why this isn't unavoidable, and that would be to introduce analog (or even analog- ish, truly near-zero) factors into the monitoring scheme. But of course, that doesn't let you use plugs in realtime on your input. <--- not a true statement, because there is not such thing as 'realtime' plug processing on input in the digital domain. So yeah, it's great that you can put XYZ plug with the über-fancy GUI on an input. Yeah, man, real cool! Look at that thing! It sounds sweet! Let's rock! (etc.) Only thing is, it'll throw your timing off (in addition to software monitoring latency). The bells and whistles and false claims cause people problems. That's my problem with the whole thing.

Look, I'm not advocating turning back the clock to all-analog electronics. Really, I'm not a Luddite by any stretch. But when it comes to musician's timing, there is no better monitoring scheme in the world than an analog monitoring signal path -- the only one that doesn't introduce the "do I suck all of a sudden?" phenomenon.

Message was edited by: iSchwartz

iSchwartz wrote:

But how is it possible to hear the sound earlier when the A/D and DAW-induced latency will inevitably make it later?

If a guitarist is standing up, his ears are probably three to five feet from the amp. That means he will hear the sound after about three to five milliseconds.

If the mic is six inches from the amp, that's half a millisecond. Add the 1.6 ms of latency and the guitarist hears the sound after about 2-2.5 ms.

It's possible because since the mic is closer to the speakers than his ears, even after a latency delay, the sound still gets to the headphones faster.

Since you're putting the mic next to the instrument and not next to the player's ears, you can end up with a delay shorter than the acoustic latency. The mic is in a different place so the two sound paths are independent - the headphone sound doesn't include the delay from the instrument to the player's ear, so the digital latency isn't added to the time of the sound traveling to the ear. They are parallel paths.

As long as the digital latency is shorter than the acoustic latency, I don't see why "unnatural" latency is any worse. A guitarist who has his amp more than two feet from his head is always going to hear more than 2 ms of delay between playing and hearing. I don't understand why, if that doesn't cause problems, why a 2 ms delay caused by digital latency would cause problems.

that's true and brilliant, but we also see examples of DSP accelerated plugins that are able to be part of signal path at close to, if not analog speeds. what are your thoughts on that?
I have not used protools HD for example, so I don't have any direct experience with such systems, but to me it seems a little strange. It is absolutely right that FFT requires bins filled with samples (buffers) to be able to perform, but how is this different when it is being performed on DSP chips versus native power in the box.
what we had been talking about up to this point is simply the audio signal path only without plugs, but of course every time you add a plugin you add latency (well plugs that perform FFT related tasks), but on a native system you introduce latency doubly because every time you add a plugin you introduce latency, and also the more plugs you add the less processing power is available and therefore tasks take longer which is another cause of latency on the scheduling level.
I assume that advancement in parallel processing may help alleviate this process and help eliminate that second type of latency, but not the first.
any ideas?

OK, I see your point. But the scenario you described is one of only several where latency may or may not be a problem. Variations:

1• guitar player is in the studio close to the amp, but has cans on
2• guitar player is in the studio away from the amp (to get the room vibe) but has cans on
3• guitar player is in the control room monitoring live through the speakers
4• guitar player is in the control room with cans on
5• guitar player is on stage with in-ears
6• guitar player is on stage listening live to his amp

Each of the above represents a scenario where 'unnatural latency' may or may not prove to be a liability. The context has to be taken into consideration.

To answer a couple of earlier questions:

- In Pro Tools TDM the latency is essentially defined by the converters.

- You can get essentially no monitoring latency (almost the speed of light) with an analog signal. Lots of interfaces do that: Digidesign M-Box, MOTU 828 (original), and many cheaper interfaces. Sure analog equipment can have some phase shift, but it's a non-issue.

And to underline an important point: _everyone here is going on about monitoring latency with live recording. The more important latency figure, at least to people like me, is how long it takes to play sounds using Logic and other V.I.s.

no, in the case of modern low latency systems, the buffer time is the upper limit allowed for all tasks to be performed. in logic low latency mode, any plug-ins that report needing more time than is allowed will be bypassed. however, the lower the buffer size the more cpu power required (more calculations per cycle). the point is that modern computers are able to handle all of this in the time it takes sound to travel 50cm.

i would regard that as so insignificant as to be in the realm of the difference between recording on a hot day and recording on cold day (sound travels at different speeds dependent on air temperature) or up a mountain or down a mountain, or a rainy day or a sunny day (sound travels at different speeds depending on air pressure).

guys,
this is getting a little too irrelevant because all these latencies you're talking about between the guitar player and guitar are only part of what we call a reverb. sound in these instances is not direct, but the collection of reflections, which is in a way a collection of latencies. so your ear doesn't care if it hears it 3ms or 5 ms from the guitar amp.
this is different than monitoring a track directly through your headphone and you hearing the latencies between when you play the guitar and re-hear it in your headphones.
that is also why in big concerts the guitar players will have monitors in ear, not only to hear himslef clearly but because of delay. in fact they use delay units to delay signals between stacks of speakers in such large venues because the cable runs (analog) are so huge that they actually introduce latencies higher than 10 ms which are certainly perceived by our ears as delay.
so, in actual practice delays of less than 7 ms or so are not really perceived by us as a delay in the signal. so if apogee are able to provide even 4 ms latency using native processing, I would find it amazing. of course no point to say they can do that when you have to run at 96k and only be able to run 8 tracks simultaneously !!
the technology is moving forward and I think that what we perceive of as DSP processing technology will become a common place in our studios and allow for more analog like work flow with benefits of digital.
for starters parallel processing needs to catch up for us to be able to fully utilize our current processing powers, on top of that there is a technology what will allow us to distribute processing power to powerful graphic cards in our machines, which is much the same approach as having DSP. I think that will have great implications for us in the Audio world (supposedly we will see that in Snow leopard) on top of that the coast of technology is always decreasing and so DSP chips are cheaper to make and will be used more prominently to have systems like the protools HD running for a lot less money and with any software/hardware combination. it is a matter of time.

Always kinda amazes me how far people will go to argue with something that is easily confirmed or rejected through experiment.

The position that I (and Jim Frazier, and others) am taking is very simple: I've monitored myself through both digital and analog systems, and I prefer analog. Analog tends to give me a more accurate sense of tone and timing. I've tried both, and that's what I hear. Simple as that.

So, in response to that, we get a ton of rationalization about how complex and messy music and recording are in the first place, so how could this matter and how could that matter... where does any of that lead ?

Digital monitoring introduces clearly-measurable latency, while analog does not. That's just a fact. And during a real session, that latency either bothers you or it doesn't.

There's not a lot of use for conjecture here. We're talking about music. No one cares whether or not 1ms of delay in the headphones should matter. We can come up with dozens of reasons why it should be "insignificant," but the only thing that really matters is whether or not it does matter, i.e. do we notice the additional latency or not, and if we notice it, is it a good or bad thing ?

And as generally happens in these sort of discussions, *no one* is jumping to say, "I've listened and I can't hear a difference."

You can't argue with ears, folks. We're talking about music here. Either you hear something or you don't. Everything else is just talk.

James

hmm i think you need to read my posts a little more carefully - unless you were not directing that post at me?

i have already said that i have done both as well and really it has made no difference to me. i am not arguing about 'sound' only the effectiveness or relevance of latency in this scenario.

later i went on to say that having tried to patch in my reverb directly to my interface inputs, i was not convinced of the sound. now, talking about sound is different matter, and even with my extremely basic setup, i really struggle to get my ITB mixes to sound as satisfying as my OTB mixes. i have done blind A/B tests with colleagues that confirm this. yet, i do see the way forward as being ITB and i persevere trying to get them to the same quality as my OTB mixes, largely because i still think my OTB mixes have room to improve and i will only be able to achieve easily and practically ITB in the future.

if it is purely down to sound, you have a different argument. if it is down to with latency at those minute amounts makes an iota of difference, then we have an argument - and actually jim felt pretty much the same way.

in fact he makes a point that kind of confirms what mike and i have been trying to say:

he was saying that by software monitoring in the past he found ultimately he wasn't satisfied with his results. this was probably due to latency, but latency a great deal more than the amounts we are talking about here. but it was not immediately noticeable - he couldn't tell the difference straight away, even though he was probably dealing with latency that goes beyond what is not going to affect performance.

no one disputes that latency is an issue - it has to be gotten down to a level where it is transparent and has no bearing on the performance. in the past, at least natively that was not entirely possible - close, but no cigar. NOW, with latencies the same as PT HD systems, it IS possible, and is a viable alternative to a PT system and a fraction of the cost with absolutely first class gear and first class sound. break out the cigars.

choosing a analogue system over a symphony or for that matter a PT system should not be based on latency. it is so small it is a non-issue. choose hardware perhaps because you prefer the sound of hot transistors to play to, or because it is cheaper and the flexibility is not an issue for you. choose it because of cost - analogue hardware monitoring is still the cheapest way.

but don't say 'well 1.6ms is so much latency you could drive a truck through it and its going to ruin everyones performances or at least put them off' because i honestly think that is guff.

Well, since the guff-like "drive a truck through it" remark was mine, guess who feels obliged to chime in? 😉

Please remember that this 1.6ms latency spec is unrealistic in the sense that you have to run at 96 to achieve it. And even David has said that it taxes your computer, lowers the track count, etc. So your argument that it's all become economical can only be reflected in pounds/dollars/yen/whatever +but not performance+. Therefore it's a false economy, and while the technical claim holds water, it's not practical for most normal recording applications. It's like MPG ratings for cars. 32 MPG highway, but realistically you get about 12 in traffic. It's no different with latency specs.

So maybe you're right. You can't hear 1.6 ms. But when you run that system at a more realistic 48K,
I g-u-a-r-a-n-t-e-e you'll hear it when the latency becomes closer to 4 ms.

I don't think it's time to celebrate technology until it meets or exceeds the performance of the gear it's attempting to replace or at least replicate.

There has to be a point where we realize the importance of respecting the immediacy of a musician's reaction to sound. I don't think that any arbitrary number of milliseconds can be thrown out willy-nilly to justify technology that doesn't perform up to the very benchmark used to rank its performance -- the analog signal path.

BTW, The Guff is the place where unborn souls are kept. I'll bet it's a pretty big place, probably big enough to drive a truck through...

😉

Well, since the guff-like "drive a truck through it" remark was mine, guess who feels obliged to chime in?

er....was it?....i don't recall.....ahem...sorry...

but

g-u-a-r-a-n-t-e-e you'll hear it when the latency becomes closer to 4 ms.

again i am deeply skeptical about. but let's look at the real world scenario...

the whole point of monitoring through software is to use software plug-ins. very likely a touch of reverb. how is the ear supposed to be able to distinguish between the latency of a dry signal against the background of reverb that is creating by its nature a whole bunch of delays? 4ms isn't enough time for middle C to finish a cycle. certainly it is still so small as to make no difference.

and the 1.6ms is realistic....look at the figures posted by apogee...also my friend who has the system and tested it out would verify this. it works really well at 32 sample buffer and you can run plenty of tracks and record and play plenty. 32 tracks being recorded while 32 tracks is plenty. its a realistic and valid way of doing things these days with these super fast computers.

and the sharp end of the business, engineers are more than happy to record at 96k. and this is meant to be a professional solution, a native alternative to a PT rig. latency WAS an issue before but now it isn't.

but 4ms will be more than adequate for most situations anyway, so you can run 44.1 and no one will bat an eyelid. its 40ms that you start having a point and jim found that ultimately that WAS too much.

I don't think it's time to celebrate technology until it meets or exceeds the performance of the gear it's attempting to replace or at least replicate.

well it has. PT HD system was the benchmark and it can meet and maybe even just pip it. so cigars for symphony. PT HD system trying to beat or equal analogue...well fine...it hasn't, but really it hasn't made any difference, its still used in that way in a vast majority of studios.

We got an apogee symphony system and hopefully will be hooking it up soon. I'm looking forward to putting it through its paces and seeing what it can do both at 44 and 96.

iSchwartz wrote:
So maybe you're right. You can't hear 1.6 ms...

For the record, can you hear 1.6 ms? Have you tracked with 1.6 ms latency and been able to hear the difference between that and analog monitoring?

You obviously haven't been happy with monitoring with digital latency, what's the lowest you have tracked at?

Mike Connelly wrote:
For the record, can you hear 1.6 ms?

No, Mike, I can't. I'm just involved in this argument for the sport.

Of course I can tell!!! Do I have nothing better to do with my time than to post my views with nothing to back it up? I can only hope you can say the same thing.

But to answer your question, yes, Mike, I can tell when something is "off" as little as 27 samples. I can routinely hear when something is off by .5 millisecond. I can tell if picture and music are out of sync by 1/4 frame.

Regarding the 27 samples, how can I tell? Because it was when I arrived at the conclusion that I needed to set my recording delay to -27 samples that all of a sudden, live recordings (rhythm guitar, for example) finally started to feel exactly right against my tracks. Previously they did not. So, let's do the math: at 44.1K that equals a timing error of .61 milliseconds, just over .5 milliseconds, per my claim above.

To prove the point that 1.6 milliseconds is NOT friggin inconsequential, I propose a very simple test:

1. At 120 BPM, Record a 8th or 16th note HH track from a virtual instrument. Quantize it.
2. Duplicate that track and the region. Now you have two regions playing the same thing using the same instrument
3. Set your realtime delay parameter's view settings to milliseconds
4. Shift the duplicated track by 3 delay increments. _That's 1.5 milliseconds._

Now, play back the track. If you can't hear that the two tracks are now at least phasing or that there's a teensy-tiny flamming between attacks, then you're right. You can't hear 1.5 milliseconds. But if you do (and I daresay you will), I have proven my point.

sounds like an excellent experiment that will put an end to all of this fantastic debate. I hope !!

The studio I work at has a 3 card symphony system with 4 DA16s and 1 AD16.

All the digital inputs (64) and outputs (16) are being used, i.e., we have machines hooked up and the Logic environment has inputs and outputs for all of them. Very rarely do we ever use any softsynths of plugins in the Logic environment - it's mostly just a mixer for lots of Giga PCs.

I find that a buffer setting of 256 is the only one that works. Strangely, going to higher latency causes as much problems (stuttering, etc.) as reducing it . . . this make sense to anyone?

48K.