Print

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

[ronkupper]

Hello Guys,

I'm debating between the following room calibration options.

Acourate is a strong candidate since it has good references here and detailed guides by the community experts.

But I am thinking that Dirac might be more suitable for me because:
1. It support multi-channel room correction out of the box (Whereas I'm not yet sure of how the process would look like in Acourate for multi-channel)

2. It works at a driver level and thus support any application with no mentions of latency or video sync issues that I could find (I hear a lot of internet audio and Acourate was rather limiting to JRiver for the best experience to my understanding).

My main interest is still high fidelity stereo music but I do have a library of multi-channel material which I am very fond of (and of course there are movies as well).

I would like to hear from your experience and perhaps gather points in favor of one software or the other.

Thanks,
Ron

[6233638]

Dirac's driver-based correction is both good and bad.
It's good because it makes setup easy, and works with all media sources played on the PC.
But at the same time, it only works for the PC's audio device.
 
Acourate works by creating a file which is used in the JRiver convolution engine, which means that it can be applied to any number of devices.
 
So rather than playback directly on your PC, you could also be using it to correct a wireless speaker somewhere else in your house for example.
 
 
I contacted Dirac about the possibility of a VST plugin, which would let you use it on any device via Media Center, and they said that they have been testing one internally, but do not know if they plan on releasing it.
If they do release a VST plugin rather than relying on their driver for playback, then it would be a lot more appealing to me.

[mwillems]

I haven't used the full versions of any of the three suites at length, but I wanted to comment on your 2) point:  any convolution based room correction solution is going to introduce latency.  The more you're trying to "do" in the convolution filter, the longer the filter will need to be in order to work correctly.  Depending on the length of the filter, you will have more or less latency, that's just how convolution works.  So even if there is a difference between Acourate and Dirac in the software latency department, the convolution filter itself is an unavoidable source of (usually quite significant) latency.  I've made a lot of different convolution filters using the free tool RePhase, and almost all of them have introduced enough latency that I couldn't use them with internet video.  

Obviously added fixed latency is irrelevant for audio, and not always relevant for video, provided that you're playing the video in JRiver.  The only exception I'm aware of is DVD disc playback, which is limited by microsoft's DVD navigator.  This can be gotten around by ripping to mkv.  

So any of the convolution based solutions will likely make it challenging to use them with internet video. If your only concern is about routing internet audio through JRiver, JRiver offers a software loopback function that will allow you to funnel all system sound through JRiver.  Sorry I can't be more help on the specifics of the suites themselves, but there are plenty of folks here who can weigh in on the pluses and minuses.

It's still a work in progress, but if you enjoy tweaking, I just wrote up a brief introductory guide to using free software for designing speaker/room correction in JRiver, here: http://yabb.jriver.com/interact/index.php?topic=87538.0.  JRiver's built in PEQ can't do everything convolution can do, but it can do a lot, and with much less latency.  If you want to try out convolution before paying for one of those software suites, there are some free convolution filter design tools out there.

[mattkhan]

Hello Guys,

I'm debating between the following room calibration options.

Acourate is a strong candidate since it has good references here and detailed guides by the community experts.

But I am thinking that Dirac might be more suitable for me because:
1. It support multi-channel room correction out of the box (Whereas I'm not yet sure of how the process would look like in Acourate for multi-channel)

2. It works at a driver level and thus support any application with no mentions of latency or video sync issues that I could find (I hear a lot of internet audio and Acourate was rather limiting to JRiver for the best experience to my understanding).

My main interest is still high fidelity stereo music but I do have a library of multi-channel material which I am very fond of (and of course there are movies as well).

I would like to hear from your experience and perhaps gather points in favor of one software or the other.

Thanks,
Ron

There is a description of one process for multichannel on acourate in the acourate yahoo group (registration required), it boils down to treating it as a set of stereo channels (l+r, l+c, l+sr etc).

[ronkupper]

But at the same time, it only works for the PC's audio device.

The HTPC is the sole device I use for both audio and video so that is less of a concern for me.

Regarding the VST plugin - are you referring to something like what Acourate has or simply for the FIR filters using JRiver?
I see a big advantage in this in regard that it would eliminate any video syncing issues entirely.

I haven't used the full versions of any of the three suites at length, but I wanted to comment on your 2) point:  any convolution based room correction solution is going to introduce latency.  The more you're trying to "do" in the convolution filter, the longer the filter will need to be in order to work correctly. 

Understanding this could be an issue I've searched the web for latency or video syncing "complaints" regarding Dirac and couldn't find any. Perhaps it is because people use it primarily for audio.

Anyway - Imbedded in your respond is one of my main concerns that "Maybe there is less latency with Dirac because it is doing less (sonically) then Acourate...?"

Is there a comparison of filters and methods used between the products?

There is a description of one process for multichannel on acourate in the acourate yahoo group (registration required), it boils down to treating it as a set of stereo channels (l+r, l+c, l+sr etc).

Thanks a lot, I'll register and have a look. Have you tried it successfully?


Another thing -
For Dirac it is stated (recommended even) that one can use a USB based condenser microphone for measurements. Is it also possible with Acourate considering its ASIO requirements?


Thanks again,
Ron

[mwillems]

Understanding this could be an issue I've searched the web for latency or video syncing "complaints" regarding Dirac and couldn't find any. Perhaps it is because people use it primarily for audio.

Anyway - Imbedded in your respond is one of my main concerns that "Maybe there is less latency with Dirac because it is doing less (sonically) then Acourate...?"

I don't think Dirac is doing less, they may be doing it differently than their competitors.  They claim their reduced latency is due to a proprietary hybrid filtering method that isn't fully IIR or FIR.  If that's the case, they may just be using low latency IIR filtering as far as it will take them, and then "finishing" with FIR filters.  That's a good idea because it would substantially reduce the latency required, but I'm not sure if you could push the latency far enough down even so without offering an option to turn off the FIR for latency sensitive applications.  It's theoretically possible that Dirac is actually "ahead" in this respect and lipsync may work fine with Dirac, but if I were you I'd confirm that it really works that way before buying (they offer a free trial).  

Realistically, using FIR filters to do all your correction (like Acourate or Audiolense do) requires a lot of latency, and even using it for more limited correction (which may be the case with Dirac) requires potentially significant latency.  Typically, the filter needs to be long enough to handle the period of the frequency you're adjusting, and a complex filter might need to be quite a few periods.  The period of a 20Hz tone is about 50ms, so even making a filter that runs a single period at 20Hz is more than enough to blow lipsync completely.  20Hz is a worst case scenario, but with significant adjustments below 100Hz, you can quickly get into triple digits of milliseconds worth of latency for the filter alone.  That kind of latency dwarfs any typical software latency by an order of magnitude.  If Dirac can push it down to 10 or 15ms, they might be able to keep internet video lipsync working correctly, but I don't know how they could accomplish much FIR correction below 100Hz that way.

My recommendation would be to E-mail them and ask them about it, or (since they offer a trial) try it and see.

[mojave]

I use Audiolense and really like it. It is so easy to do a multi-channel measurement and apply filters. For music I use very long filters. For movies or TV I use shorter filters to make it easier to skip forward/backward with an awkward delay. For other computer sources (ie. internet), I use Audiolense's Minimum Delay XO filters. They add only 15 ms of latency. There is no noticeable lipsync latency. I use ZoneSwitch to automatically switch between filters/zones.

The output latency of my Lynx AES16e is less than 1ms. I recently tested a Solid State Logic MadiXtreme card and Alpha Link MX DAC. Its round trip (analog out/analog in) latency is .9ms per testing by SoundOnSound. For just the output it would probably be less than .5 ms. I also used it with computer audio routed through JRiver and didn't notice any latency. I've found that PCIe cards for digital output have less latency than Firewire or USB. However, I don't think the difference is enough to make lipsync an issue when going with Firewire/USB.

[mwillems]

I use Audiolense and really like it. It is so easy to do a multi-channel measurement and apply filters. For music I use very long filters. For movies or TV I use shorter filters to make it easier to skip forward/backward with an awkward delay. For other computer sources (ie. internet), I use Audiolense's Minimum Delay XO filters. They add only 15 ms of latency. There is no noticeable lipsync latency. I use ZoneSwitch to automatically switch between filters/zones.

Just out of curiosity, how does the "minimum delay" filter compare (measurement-wise) to the full length filter?  It makes sense that they'd offer a "low latency" option, I'm just curious how much daylight there is between it and the full length filters in terms of how the correction looks (especially on the low end).  

Everytime you post about it, I get a little closer to buying Audiolense  

[mojave]

Just out of curiosity, how does the "minimum delay" filter compare (measurement-wise) to the full length filter?  It makes sense that they'd offer a "low latency" option, I'm just curious how much daylight there is between it and the full length filters in terms of how the correction looks (especially on the low end).

After I first determined that Audiolense's simulated result matches a real measurement fairly close, I haven't taken the time to actually do a real measurement with the convolution filters in place. I guess I've been lazy.

Here is what the difference looks like with no smoothing on the simulation. The first chart is the Minimum Delay XO and is actually smoother looking. The second chart is the True Time Delay filter with a latency of 1,370 ms. Both correct the frequency response almost the same.

[dean70]

I thought the JRiver Convolution engine takes into account the latency? I have not had to do anything about lipsync since enabling filters for video  ?

[mwillems]

I thought the JRiver Convolution engine takes into account the latency? I have not had to do anything about lipsync since enabling filters for video  ?

We're talking specifically about using JRiver's loopback feature for audio from internet video played outside of JRiver (or internet video that bypasses JRiver's audio engine even inside JRiver, like netflix).  For video that uses JRiver's audio engine there should be no issues except when playing back from a physical DVD disc, and then only when the latency gets too severe.

[mattkhan]

Thanks a lot, I'll register and have a look. Have you tried it successfully?

Not yet no, I do now have my 2.1 setup working nicely so I just need to get the time to extend it to 5.1.

I thought, possibly somewhat simplistically, the extremely high latency typically involved in these filters was driven by the use of linear phase FIR filters so the lower delay filters are minimum phase instead.

[mwillems]

I thought, possibly somewhat simplistically, the extremely high latency typically involved in these filters was driven by the use of linear phase FIR filters so the lower delay filters are minimum phase instead.

I think that's definitely part of the story with Dirac; that's what I meant above about them potentially using IIR filters as far as they go. IIR filters are minimum phase filters with low latency, they're what JRiver's PEQ filters are.  

The issue though is that IIR filters work great for correcting minimum phase issues, but you can't really do certain kinds of time-domain correction/excess phase correction with conventional IIR/minimum phase filters.  In order to do time-domain correction it requires, well, a little time  .  The less you need to do (and the higher the frequency at which you plan to do it), the less time you need.

In Dirac's case, since they're running an audio driver, they could have different filter stages, they could be doing almost anything in there. It's easy to imagine they could just "turn off" the FIR under some circumstances.  What I don't fully grasp is how the all-convolution solutions (acourate/audiolense) manage to do low frequency correction at low latency; my understanding (which I'm happy to admit is incomplete) is that once you're convolving you're effectively emulating an IIR filter in FIR and you can't accomplish the same results with the same kind of low latency that a normal IIR filter can achieve.  

That's certainly been my experimental result dialing in the same filters in REW or RePhase, as opposed to directly dialing them into JRiver's PEQ: a given set of minimum phase filters turned into a convolution filter from REW or Rephase add significantly more latency to get the same level of accuracy as the exact same filters dialed into JRiver's PEQ.

[mattkhan]

What I don't fully grasp is how the all-convolution solutions (acourate/audiolense) manage to do low frequency correction at low latency; my understanding (which I'm happy to admit is incomplete) is that once you're convolving you're effectively emulating an IIR filter in FIR and you can't accomplish the same results with the same kind of low latency that a normal IIR filter can achieve. 

That's certainly been my experimental result dialing in the same filters in REW or RePhase, as opposed to directly dialing them into JRiver's PEQ: a given set of minimum phase filters turned into a convolution filter from REW or Rephase add significantly more latency to get the same level of accuracy as the exact same filters dialed into JRiver's PEQ.

I don't think an FIR is emulating an IIR as such, they are just 2 different types of filter with their own sets of pros and cons. The minidsp site has a basic overview, a slightly more in depth one at dspguru (and obviously you can google iir vs fir to your hearts content  ).

I would think that the low latency filters produced by acourate/audiolense are just minimum phase FIR filters, perhaps with fewer taps to boot. This means no phase correction & potentially less accurate frequency response correction but lower latency and perhaps a good enough amplitude correction.

It might be interesting to see some comparisons of the phase effects in mojave's earlier charts.

[mwillems]

I don't think an FIR is emulating an IIR as such, they are just 2 different types of filter with their own sets of pros and cons. The minidsp site has a basic overview, a slightly more in depth one at dspguru (and obviously you can google iir vs fir to your hearts content   ).

Sorry, I think I've confused things, and I probably shouldn't have used the word "emulation."  What I was trying to get at is that, as I understand it, performing minimum phase EQ in a FIR filter will not be as efficient as performing the identical minimum phase EQ using an IIR filter:

The main advantage digital IIR filters have over FIR filters is their efficiency in implementation, in order to meet a specification in terms of passband, stopband, ripple, and/or roll-off. Such a set of specifications can be accomplished with a lower order [...] IIR filter than would be required for an FIR filter meeting the same requirements. If implemented in a signal processor, this implies a correspondingly fewer number of calculations per time step; the computational savings is often of a rather large factor.

[from the IIR wiki page]

I didn't mean that it was literally attempting to emulate the exact processing steps, more that it was attempting to create the same minimum phase result, but that it required more time and computational power to accomplish it (in the same way that a NES emulator requires much better hardware to run than the NES it's emulating).  

I would think that the low latency filters produced by acourate/audiolense are just minimum phase FIR filters, perhaps with fewer taps to boot. This means no phase correction & potentially less accurate frequency response correction but lower latency and perhaps a good enough amplitude correction.

I think we're agreeing on this part. I assume acourate and audiolense are using FIR filters to create a minimum phase response; I'm not sure how they could be doing anything else when their only output is a FIR filter.  They definitely would have to reduce the taps to reduce the overall latency.  The part that I was expressing bafflement about above is that, to my understanding, there are certain minimum latencies involved in doing any kind of FIR manipulation (even manipulation of frequency response), and that those minimum latencies increase as the frequency of interest falls.  So what I'm not quite sure about is how they manage to do any meaningful frequency correction below, say, 60 Hz with a 10 or 15ms FIR filter.  Maybe the answer is, as you say, it's "good enough" correction, or maybe the limitations of FIR latency are not where I think they are  

The distinction I was making about Dirac is that it's not necessarily limited to using only FIR filters, because they're running their own audio driver.  So they could have an IIR stage, and then an FIR stage and get the best of both worlds in terms of latency and flexibility.  But to be clear, I have no idea what Dirac actually does.  Their literature just says their method is not fully IIR or FIR, and so I made an inference.

It might be interesting to see some comparisons of the phase effects in mojave's earlier charts.

100% agreed, I would expect to see a much greater divergence there.

[mattkhan]

Right I understand now, I was being rather pedantic for a change

From what I've read, the latency through a (linear phase) FIR filter is 1/2 the no of taps in the filter (in samples) so a 65k tap filter (which IIRC acourate uses by default) will have a delay of ~32k samples or about 0.66s at 48kHz which seems about right for my setup. I have no idea what the algorithms are that reduce latency for a minimum phase filter though, the maths is beyond me. A DSP expert needs to step forward to answer that one.

[mwillems]

Right I understand now, I was being rather pedantic for a change

From what I've read, the latency through a (linear phase) FIR filter is 1/2 the no of taps in the filter (in samples) so a 65k tap filter (which IIRC acourate uses by default) will have a delay of ~32k samples or about 0.66s at 48kHz which seems about right for my setup. I have no idea what the algorithms are that reduce latency for a minimum phase filter though, the maths is beyond me. A DSP expert needs to step forward to answer that one.

Yeah that's also my understanding of the taps/latency connection.  The question from my perspective (for our hypothetical DSP expert) is: how many taps do you need to do significant (i.e. more than a dB or two) full band minimum phase frequency correction?  

Until our expert shows, you can try an experiment with RePhase if you want to see the phenomenon I'm describing.  RePhase allows the user to dial in minimum phase EQ in a FIR filter, and allows you to specify the length of that filter in taps.  You can get the filter length almost arbitrarily short (I think the minimum taps RePhase will allow is in the hundreds or low thousands).  

Try dialing in a minimum phase parametric boost of, say 6dB at 20Hz, and then try creating that filter.  RePhase will model the correction it thinks it can achieve with the given filter length and show it to you superimposed over the target.  So you can plug around with different numbers of taps and see what the effects are.  My experience when I tried it last (admittedly with an older version of RePhase) was that, when the number of taps got small enough and the filter was low enough in frequency, the filter stopped being ideal, and eventually stopped really "working" at all.  

That seemed, to me, consistent with the "minimum latency" idea, but realistically, I recognize that I may just have identified the limits of RePhase's optimization algorithms (or my understanding of RePhase) and not the limits of FIR filtering generally.  

Audiolense's minimum latency filters were around 15ms according to mojave's illustration, which at 44100KHz, is probably around 1200 taps.  That's not a lot of taps; but based on mojave's post above, audiolense seems to be accomplishing frequency correction pretty successfully and pretty far down (unless mojave's system is just naturally flat to 8Hz, which is possible ).  

[mattkhan]

I think it is the low frequency side of things that consumes taps. AIUI more taps means more detail (sharper/steeper slopes are possible) and the no of taps required to maintain the same detail increases as the frequency affected decreases. Therefore if Mojave has a naturally flattish response in the sub area (perhaps he has >1 sub nicely positioned) then a perfectly good job could still be done with the reduced bandwidth of the filters.

I know some systems deal with the problem by breaking up the frequency range into smaller buckets so that resources can be targeted at problem areas more efficiently and the resulting filter then stitched back together.

[mwillems]

(perhaps he [Mojave] has >1 sub nicely positioned)

I think he just might   http://yabb.jriver.com/interact/index.php?topic=77075.0

[mattkhan]

I think he just might   http://yabb.jriver.com/interact/index.php?topic=77075.0

ha, yes I think that counts!

FWIW there is a thread on the acourate yahoo group entitled "Making a short filter" which has some details on using acourate for this purpose, lots of flexibility but you clearly need to know what you're doing. It does include this comment on linear phase filter resolution with reference to a 44.1kHz linear phase filter with 65536 (~0.75s) vs 2048 (~23ms) taps;

The filter resolution is 0.673 Hz in the first example but only 21.53 Hz in the 2nd example. This is just 5 frequency bins below 100 Hz.

clearly you're not going to get v far with linear phase at sub frequencies when you have a small latency budget unless you have a pretty naturally flat response already.

The thread also has lots of detail on approaches for how to go about this in acourate. I would go so far as to say this thread sums up acourate, it gives you *all* of the tools but no hand holding in the app itself. You need to know what you're doing in order to do it. The support on offer in the user group and from Uli is v comprehensive though.

I would therefore suggest, to the OP, that choosing acourate means choosing to learn more about DSP than perhaps you ever thought you would learn. This can be a good or a bad thing depending on your point of view

Note that I'm referring to going off piste here, acourate has a set of macros that deal with the common case of "give me the highest quality you can and darn the latency" v well (albeit that still needs some hand holding to get the most out of it 1st time round).

[mwillems]

FWIW there is a thread on the acourate yahoo group entitled "Making a short filter" which has some details on using acourate for this purpose, lots of flexibility but you clearly need to know what you're doing. It does include this comment on linear phase filter resolution with reference to a 44.1kHz linear phase filter with 65536 vs 2048 taps;

The filter resolution is 0.673 Hz in the first example but only 21.53 Hz in the 2nd example. This is just 5 frequency bins below 100 Hz.

clearly you're not going to get v far with linear phase at sub frequencies when you have a small latency budget unless you have a pretty naturally flat response already.

That confirms my suspicions. And in that case they're talking about a 2048 tap filter.  23 Milliseconds is pushing it for lipsync (<22 is the film standard), especially when accounting for the fact that there's other latency in the system besides the filter.  To get down into the 10 to 15 millisecond range (800-1200 taps or so), you're talking about a resolution of more like 40-60Hz, which means it would be hard to do much of anything with a filter that short in the sub-bass range.  

[mattkhan]

That confirms my suspicions. And in that case they're talking about a 2048 tap filter, which is about 25 milliseconds at 44.1.  25 Milliseconds is pushing it for lipsync, especially when accounting for the fact that there's other latency in the system besides the filter.  To get down into the 10 to 15 millisecond range (800-1200 taps or so), you're talking about a resolution of more like 40-60Hz, which means it would be hard to do much of anything in the sub-bass range.  

bear in mind this is talking about linear phase only, the alternative is to switch to a minimum phase setup which will be faster (I believe the latency in this will vary with the filter) but has an impact on phase and hence needs to be designed to the situation. There is also an option in acourate to convert a set of linear phase filters to minimum phase, if I get a chance I will give that a whirl and see what comes out.

[mwillems]

bear in mind this is talking about linear phase only, the alternative is to switch to a minimum phase setup which will be faster (I believe the latency in this will vary with the filter) but has an impact on phase and hence needs to be designed to the situation. There is also an option in acourate to convert a set of linear phase filters to minimum phase, if I get a chance I will give that a whirl and see what comes out.

If you have the time, I'd be very interested to hear about the results.  For instance, I'd love to know what the comparative latency is of a given minimum phase filter made with acourate, compared to the same minimum phase filter modelled using an IIR filter (like the one's in JRiver's PEQ).  As noted above, I've looked at the comparative latency of minimum phase filters made with RePhase (for example) as compared to IIR filters, and they did not compare favorably. I'm very curious if Acourate has "cracked the code" on that issue, so to speak.

[mattkhan]

If you have the time, I'd be very interested to hear about the results.  For instance, I'd love to know what the comparative latency is of a given minimum phase filter made with acourate, compared to the same minimum phase filter modelled using an IIR filter (like the one's in JRiver's PEQ).  As noted above, I've looked at the comparative latency of minimum phase filters made with RePhase (for example) as compared to IIR filters, and they did not compare favorably. I'm very curious if Acourate has "cracked the code" on that issue, so to speak.

AIUI an FIR filter is typically an order of magnitude slower than a comparable IIR (and that's just the way it is).

Does jriver report the latency of a convolution filter anywhere?

[mwillems]

AIUI an FIR filter is typically an order of magnitude slower than a comparable IIR (and that's just the way it is).

That's my understanding as well.  I'm just not sure how much slower it "has" to be.

Does jriver report the latency of a convolution filter anywhere?

Not really.  RePhase claims to tell you how much latency the filter is going to add when you generate it, but that estimate isn't always quite right.  When I did my comparative tests, I used the time lock feature in Holm to take two measurements and compare the offset of the impulses which is the test I used that convinced me that doing REW correction as an exported WAV instead of copying over into PEQ was a "bad deal" in terms of latency.

[mattkhan]

Not really.  RePhase claims to tell you how much latency the filter is going to add when you generate it, but that estimate isn't always quite right.  When I did my comparative tests, I used the time lock feature in Holm to take two measurements and compare the offset of the impulses which is how I found out that doing REW correction as an exported WAV instead of copying over into PEQ was a "bad deal" in terms of latency.

I think I'd have to do that too then. I was also going to try routing my timing reference (in REW) through the R channel so the timing reference gets convolved as well. I suspect this won't work but has to be worth a try (as sending one through convolution and the other not gets it completely confused).

[mattkhan]

If you have the time, I'd be very interested to hear about the results.  For instance, I'd love to know what the comparative latency is of a given minimum phase filter made with acourate, compared to the same minimum phase filter modelled using an IIR filter (like the one's in JRiver's PEQ).  As noted above, I've looked at the comparative latency of minimum phase filters made with RePhase (for example) as compared to IIR filters, and they did not compare favorably. I'm very curious if Acourate has "cracked the code" on that issue, so to speak.

FWIW today I learned that acourate has a "prefilter" feature which allows you to specify arbitrary IIRs that get embedded into the final correction but are not subject to the windowing etc done by the room macros. I still haven't had time to try out the minimum phase approach, still twiddling knobs on my primary setup before I get time to do that.

[mwillems]

FWIW today I learned that acourate has a "prefilter" feature which allows you to specify arbitrary IIRs that get embedded into the final correction but are not subject to the windowing etc done by the room macros. I still haven't had time to try out the minimum phase approach, still twiddling knobs on my primary setup before I get time to do that.

No worries; that's interesting information by itself.  I'll be curious to hear your thoughts once you've gotten everything set up.

[mattkhan]

No worries; that's interesting information by itself.  I'll be curious to hear your thoughts once you've gotten everything set up.

slow going with the setup in the Khan household (pesky work getting in the way) but FWIW the next version of acourate has a feature that addresses exactly this requirement for the OpenDRC platform, i.e. it will produce a set of biquad filters to deal with the LF correction alongside a (I think 6k tap) phase correcting FIR for HF. This means it should fit inside a much tighter horsepower/latency budget but still give good results.

[Flak]

As correctly stated before Dirac Live uses a proprietary structure of both FIR and IIR filters.

The latency introduced by a Dirac Live correction can be seen by looking at the distance in milliseconds between the before and the after correction pulses in the DLCT screen (it shows the average of the nine different positions that define the listening area) and it can be something like 15/16 ms:



There is switch to select between "max performance" and "min latency" so that an even lower latency can be obtained, but at the cost of a lower performance also as you can expect.

Ciao, Flavio

[zydeco]

I'm really interested to understand any progress on low latency audio processing. My own set-up involves (acourate) filters for two-way active stereo speakers and then 4 mini subwoofers - a total of 8 channels. All this is working and, when I've got time to measure / improve, it gets better. The set-up is a joint A/V system which has been fine until just recently when Netflix arrived in Australia and became a family favourite. It's made me wonder if this path is wrong given the evolution towards A/V streaming given that <0.2ms latency seems very difficult.. If I understand things, then reducing the no. of channels (e.g., feeding subs with full-range signal and using on-board EQ) won't be much of a benefit as the mains would still have the long delays. The only simple answer looks to be to create some simple parametric eq. for A/V but I've not really given much thought as to how this would work. Any suggestions would be appreciated and, meanwhile, I'll re-read this thread.

[mwillems]

I'm really interested to understand any progress on low latency audio processing. My own set-up involves (acourate) filters for two-way active stereo speakers and then 4 mini subwoofers - a total of 8 channels. All this is working and, when I've got time to measure / improve, it gets better. The set-up is a joint A/V system which has been fine until just recently when Netflix arrived in Australia and became a family favourite. It's made me wonder if this path is wrong given the evolution towards A/V streaming given that <0.2ms latency seems very difficult.. If I understand things, then reducing the no. of channels (e.g., feeding subs with full-range signal and using on-board EQ) won't be much of a benefit as the mains would still have the long delays. The only simple answer looks to be to create some simple parametric eq. for A/V but I've not really given much thought as to how this would work. Any suggestions would be appreciated and, meanwhile, I'll re-read this thread.

This thread gives a thorough semi-automated solution for creating a bank of parametric EQ filters based on measurements using free software: https://yabb.jriver.com/interact/index.php?topic=87538.0 .  You'll need to go a bit beyond what's in the guide for crossovers/subs, but the basic principles are the same.

It won't do everything that Dirac does, but it will get you a 90% solution with virtually no latency.  You can have an arbitrary number of IIR filters (like the ones JRiver's parametric EQ uses) with virtually no measurable latency cost.  I routinely run banks with 80 or 100 filters and I the added latency isn't measurable with home equipment (i.e. the random fluctuations in my latency measurements are larger than the added latency from PEQ).

[zydeco]

This thread gives a thorough semi-automated solution for creating a bank of parametric EQ filters based on measurements using free software: https://yabb.jriver.com/interact/index.php?topic=87538.0 .  You'll need to go a bit beyond what's in the guide for crossovers/subs, but the basic principles are the same.

It won't do everything that Dirac does, but it will get you a 90% solution with virtually no latency.  You can have an arbitrary number of IIR filters (like the ones JRiver's parametric EQ uses) with virtually no measurable latency cost.  I routinely run banks with 80 or 100 filters and I the added latency isn't measurable with home equipment (i.e. the random fluctuations in my latency measurements are larger than the added latency from PEQ).

Wow. That's fabulous. It'll take some time to digest but looks just the ticket.