Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[innesireinar]

Why these units do not exist (at least from my info)?
Because this technology likes to eat many CPU GHz it could be a good solution avoiding to stress our computer.
Comments?

ranieri senni

[Roland Storch]

There is one: Sony DRE-S777

Not shure but I heard also Quantec wants to build a convolution unit.

[Nika Aldrich]

There are two problems.  One is patent/copyright issues on the process used.  The other is the amount of processing power required.  Since convolution is a pure brute-force style math problem that requires very little "intelligence," everybody is in equal competition - Lexicon doesn't have anything really unique to get you to buy their box over TC, etc.

The problem is that the development cycle for the product is long enough that the technology surpasses it in the much-wider computer market.  In the 1 year it would take (at minimum) to churn out a new, hardware convolution unit the price of the PC will drop by half and become twice as powerful.  Then, when people compare what a $2k hardware reverb box can do vs. a $1k computer the manufacturer simply won't be able to sell enough to pay off the investment.  Further, as soon as someone else enters the market, whoever's box is later in development will have the edge because it will inherently be more powerful for the same money.

Yes, it would be nice if such a device existed.  In the meantime, some of my clients are setting up dedicated used G4s off of ebay with an Mbox and a copy of Altiverb for $2k and using them as dedicated processors.  Makes a lot of sense when you think about it.  I have a client who has a recital hall rigged that way - permanently - with their favorite settings in the startup folder.  They just turn it on, wait 1 minute and then pass audio through.  There's no monitor, keyboard, or mouse hooked up.

Nika

[innesireinar]

Thanks.
The story of old G4 and Altiverb could be a good trip, but what about of latency? Last month I moved from Logic 7 to PTLE 002R with a G5 2x2 and with buffer below 512 it's hard to running Altiverb. I've not understood if the guys who you're referring use Altiverb in stand alone without an host SW (if it's possible) that could be more performance because with 512 samples in+out in a host SW you will have a minimum "predelay" of 24 ms that you cannot shorten.

A question for you Nika
I've done a test to check the latency in all SW I have (Cubase SX, Logic 7 and PTLE).
To do this I've loaded a mono percussion sample in a mono track, then copy this sample to the next track with a "loop" insert (an insert only by the cable) and I've bounced this two tracks, one far left and the second far right, then I've loaded the bounced stereo split file and checked the distance between left and right (in ms). I've discovered that SX and PTLE behave in the same way and by a buffer of 512 I've measured a latency of about 25 ms (12+12 circa) that corresponds to 512/44.100=0,0116 sec (11,6 ms) whereas Logic had about 35 ms at the same buffer size setting. Is it possible that Emapple tell to users that SW are running at 512 but the true buffer size is higher?

[Bob Olhsson]

Analog reverb has more latency than any digital hardware unless you use it in bypass which is pointless!

[Joel Silverman]

There's also the Yamaha SREV1 http://www.yamahaproaudio.com/products/signal_pro/sampling/s rev1/ and the Sintefex FX8000 http://www.sintefex.com/?targ=products&src=home#SINTEFEX %20FX8000%20REPLICATOR

[bobkatz]

Bob Olhsson wrote on Fri, 07 January 2005 14:21

Analog reverb has more latency than any digital hardware unless you use it in bypass which is pointless!

Would you please explain that, Bob? I don't know what you are getting at. The extreme latency of digital reverbs like the Waves IR-1 is legendary!

[toddhooper]

bobkatz wrote on Sun, 09 January 2005 18:00

Would you please explain that, Bob? I don't know what you are getting at. The extreme latency of digital reverbs like the Waves IR-1 is legendary!

Convolution doesn't always demand huge latencies. It depends of course on the algorithm and hardware used.

If you run our TL Space convolution reverb on Pro Tools HD Accel DSP, there is less than ten samples of latency on the wet signal.

The same plugin running as native RTAS is around 500 samples of latency. General purpose CPU's aren't ideal in all respects for convolution, despite excellent price/performance for general computing tasks.

[tony666]

I believe the Behringer V-VERB PRO REV2496 uses some sort of convolution.  Why else would they charge you 300

[Daniel_Dettwiler]

Quote:

Analog reverb has more latency than any digital hardware unless you use it in bypass which is pointless

Not sure what you mean. If you mean that in a real room you have latency untill the reverb is buildet up, that is unkorrekt, because it is exactly this building up of the reverb (from the very first reflection untill the reflextions all are so many that we speak of the reverb) that is so important for the perception of "distance" and room size.

So more over if we speak of sampled reverb, we want to get it as natural as possible, and so I absolutely do not want to get latency.

General Explanation of how it works in easy words:

Now there are two possibilities of doing the convolution. One is in the "time domain", meaning that the audio that goes in the reverb-machine gets convolved straight forward with the impulse response of the sampled Room. You get almost no latency, but you need real big horse power to do that. Any computer's CPU can not do it as far as I know. The Sony DRE-777 and the Yamaha's answear to that machine can do it. I don't know for the TL Reverb on TDM.

All native versions of  Convolution Reverbs do an other method of convolution. They first do a FFT (fast fourier Transformation) of the Signal, that will bring the signal into the "frequency domain". While in the "time domain" it is relativly easy to get information about amplitude and time (look to the waveform after recording) but more difficult to get informations about frequencies, in frequency domain it is more easy to get informations about frequencies and their magnitudes (basically the amplitudes of the frequendy bands..). So once your singnal has been brought in to the representation in frequency domain, there are a lot of things that can be done very ease, that are not possible (or not easy) in timedomain (and vice versa). In frequency domain you typically have 512 or 1024 (depending on window size of the transformation) frequency bands. So if you want to make a spektral delay, it is very easy now. Or noise Reduction, just have a expander on all bands, voila. Now Convolution is also easy to do here, and it does not need so much CPU. But the Latency is generally always as big, as your window size. Meaning the more Frequency Bands you want (the better the convolution) the more latency you get. If the window size is 512, the latency will be 512 samples.

Quality wise I can not tell wheter there is any benefit to do convolution in Time domain rather then in Frequency domain. However for me alone the almost not existance of latency to me is important, thats one of the resons I use my Sony's DRE-S777.

Hope that this helps somewhat. It is not easy to explain such things in a foreign language, sorry for mistakes

Daniel Dettwiler
www.ideeundklang.com

[bobkatz]

Well, basically, what I'm getting at in terms of latency is this:


In the external hardware reverb and most typical digital reverbs, calculation latency is pretty low. This can be measured primarily by setting the unit's dry/wet setting to fully dry, and measuring the delay time from a dry signal sent into the unit. Under the presumption that the design of the unit would produce a properly timed dry signal inside the unit to its wet signal.

However, some of the convolution units, especially the Waves, produce as much as 10-12 milliseconds of latency! This means that internally, if you do a wet/dry mix, they will have to delay the dry signal by 10-12 milliseconds before mixing it with the wet.

Of course, you could mix dry/wet outside the box, and consider the additional 10-12 ms of latency as part of the "predelay" of the wet. HOWEVER, you must also consider in the early reflections (as I think they are an important part of the design of a verb) and a tetch of these can often occur in front of that 10-12 ms. So, latency is important if you are emulating a real space as much as possible.

Thus I do not understand what Bob Olhsson was getting at....  Bob O., based on what I'm talking about here, what did you mean?

[Bob Olhsson]

I'm not into "wet/dry" controls very often. I'm also not that big a fan of early reflections unless I'm trying to match replaced dialogue or something. I can see how latency might be an issue in those situations but the way I generally use reverb in pop music it just isn't because I'm almost always going to want to delay the reverberation even more. I'm also spoiled by the latency compensation in Samplitude.

[JamSync]

bobkatz wrote on Tue, 11 January 2005 00:29

Well, basically, what I'm getting at in terms of latency is this: In the external hardware reverb and most typical digital reverbs, calculation latency is pretty low. This can be measured primarily by setting the unit's dry/wet setting to fully dry, and measuring the delay time from a dry signal sent into the unit. Under the presumption that the design of the unit would produce a properly timed dry signal inside the unit to its wet signal. However, some of the convolution units, especially the Waves, produce as much as 10-12 milliseconds of latency! This means that internally, if you do a wet/dry mix, they will have to delay the dry signal by 10-12 milliseconds before mixing it with the wet. Of course, you could mix dry/wet outside the box, and consider the additional 10-12 ms of latency as part of the "predelay" of the wet. HOWEVER, you must also consider in the early reflections (as I think they are an important part of the design of a verb) and a tetch of these can often occur in front of that 10-12 ms. So, latency is important if you are emulating a real space as much as possible. Thus I do not understand what Bob Olhsson was getting at....  Bob O., based on what I'm talking about here, what did you mean?

Or you could clone the dry track, move it back equivalent to the latency (down to the sample) and take a wet feed from that. Mix with other dry track to taste.

[bobkatz]

JamSync wrote on Wed, 12 January 2005 06:52

Or you could clone the dry track, move it back equivalent to the latency (down to the sample) and take a wet feed from that. Mix with other dry track to taste.

That's so simple and elegant. Sometimes we forget the obvious. In mastering latency is rarely an issue since I can mix wet/dry in the reverb. Something that's impractical to do in most mixing situations.

I begged Waves for a proper decibel-based wet-dry ratio control in the IR-1 and we got it in the latest version! Hooray...

BK

[Jon Autry]

innesireinar wrote on Fri, 07 January 2005 10:43

Why these units do not exist (at least from my info)? Because this technology likes to eat many CPU GHz it could be a good solution avoiding to stress our computer. Comments? ranieri senni

Didn't both Sony and Yamaha have hardware convolution reverbs (in the 5 figure $$$ range, IIRC...i think it was $10-12k for the basic Sony system when it came out...) out before *any* convolution plugs were available? I remember seeing the Sony 777 one at AES before anything else...

as far as i can see, the reason the reason it isn't happening much for hardware boxes is that you can get something like Altiverb (my favorite) for $500 and i think it'd be alittle of a stretch to see a hardware box around that price. not that it couldn't be done, i just don't see it happening.

Take Care,
Jon Autry