Print

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

[jarbe]

Fantastic to see that the native convolution engine has been implementet already!

I assume that it is under development still. But being a 64 bit fp engine, how demanding (in terms of CPU) does one expect this engine to be (when it is fully developed) compared to the ConvolverVST ?


Best regards,
Jarle

[hulkss]

But being a 64 bit fp engine, how demanding (in terms of CPU) does one expect this engine to be (when it is fully developed) compared to the ConvolverVST ?

Twice as demanding. But that really does not matter if you have enough processing power. It's only a big deal if your computer is not up to it. The latency from processing is usually small compared to the latency designed into the FIR correction filters. My Intel i7 can keep up with 16 channels and 37 filter paths.

[Mikkel]

Could anyone report on performance on AMD-processors? SSE3 don't do us much good, unfortunately (a good reason to switch to Intel... as I always wanted but never found the money to do  ).

EDIT: A really spoiled feature request: possibility to use 32-bit processing instead (please don't shoot me for asking  ). For me, who is using an AMD Athlon X2 240e (2.8ghz), sometimes performance can be a problem... but then, it really is low on both heat and power consumption = quiet. One can't get it all, I guess  .


Best regards,
Mikkel

[Trumpetguy]

Could anyone report on performance on AMD-processors? SSE3 don't do us much good, unfortunately (a good reason to switch to Intel... as I always wanted but never found the money to do  ).

EDIT: A really spoiled feature request: possibility to use 32-bit processing instead (please don't shoot me for asking  ). For me, who is using an AMD Athlon X2 240e (2.8ghz), sometimes performance can be a problem... but then, it really is low on both heat and power consumption = quiet. One can't get it all, I guess  .


Best regards,
Mikkel

I am using Athlon X3, not X2, which is fully capable of doing 2 path convolution with a 132k tap filter at 15x real time. I would be surprised if the X2 shouldn't also. With many paths in a multichannel setup (in my case 15), the X3 is barely able to process a 32k filter at 2.5x real time. Thsi should give you some ballpark figure on where your X2 will perform. I am buying an i7 at my earliest convenience :-)

[RC23]

... capable of doing 2 path convolution with a 132k tap filter at 15x real time. ... able to process a 32k filter at 2.5x real time.

How do you calculate or measure the real time factor?

[Trumpetguy]

How do you calculate or measure the real time factor?

I rely on Matt. Open the convolver plugin during playback. The real time factor is displayed there.

[Mbare]

The convolution engine doesn't remove the need for manual A/V-sync - as I thought it would. I still need to delay video manually when watching movies to make it lip-synced. I'm using a 2-ch Hi-Fi with Digital Room Correction from AudioLense, no active crossovers or anything fancy. Am I missing something or do I need to start hunting for the excact video-delay once again?

[Matt]

The convolution engine doesn't remove the need for manual A/V-sync - as I thought it would. I still need to delay video manually when watching movies to make it lip-synced. I'm using a 2-ch Hi-Fi with Digital Room Correction from AudioLense, no active crossovers or anything fancy. Am I missing something or do I need to start hunting for the excact video-delay once again?

Is there a delay baked into your filters?

Could you send a copy of the configuration and filter files to matt at jriver dot com?

Thanks.

[TheLion]

Matt,

 as explained by Uli all FIR filters with phase correction have inherent delay. This is not channel delay (that comes on top of it) but delay to bring the phase in-line over the entire response. It also is a matter of "filter resolution/length" as I described earlier. That means a 65k taps 48khz filter has the same delay as a 128ktaps 96khz filter (FIR with phase correction).

[Mbare]

Wheter there is a delay or not, I don't know to be honest. I only use AudioLense to make filters and I don't know too much about the tech behind the magic. I'm sending over the files now, Matt. It might be me being stupid, of course, but I've tried with 3 or 4 movies and none are synced.

[Mbare]

...just remembered that I resample all audio to 88.2kHz (due to the Audiolense-filter). Is there a possibility that the delay is caused by resampling?

[hulkss]

I am buying an i7 at my earliest convenience :-)

Wait a few weeks for the Ivy Bridge i7. Integrated graphics HD4000. No video card needed. I am doing OK with HD3000 in a Sandy Bridge i7 but could use just a little more on the graphics side. HD4000 is a big step up.

[hulkss]

Wheter there is a delay or not, I don't know to be honest. I only use AudioLense to make filters and I don't know too much about the tech behind the magic.

If you use digital XO or True Time Domaine (TTD) correction in Audiolense, there will be delay in the filters.

[RC23]

... I still need to delay video manually when watching movies to make it lip-synced. ...

With which tool do you receive a video delay?

[taraba55]

Long time reader of this fantastic forum ,but  my first post.Before making a step in convolution I would like to "clear" the hardware side. My processor is AMD Athlon 245e which produce JMark score of 1390. I am interested in creating filters for two set ups: 2.1 channel  and 5.1 sourround (2 subwofer in serial connection) with capability to play 24/192 flac files. I have to keep my present motherboard because od ASUS Xonar ST + H6 board already in place. My question is which JRmark score is needed for this set up for optimal performance and any suggestion which of AMD processors can do that. With my present hardware in 2.1 set up can I play 24/192 flac. Thank you

gigabyte 880gm-ud2h, 8GB RAM , Seasonic 400W fainless PSU

[Mbare]

Hulkss: I use True Time Domain-correction, so that might explain the delay.

RC23: In MC, under "Playback Options", "Video", "Advanced", you can change A/V-delay; that's where I delay the video.

[jarbe]

hulkss: Upgrading to the latest version made the playback-skipping dissapear. Clearly not a CPU issue. I run 6 channels/6 paths. My Intel Atom N270 does the job very well with its 1.6GHz singel core processor.

[RC23]

hulkss: Upgrading to the latest version made the playback-skipping dissapear. Clearly not a CPU issue. I run 6 channels/6 paths. My Intel Atom N270 does the job very well with its 1.6GHz singel core processor.

Interesting info that an Intel Atom N270 with 1.6GHz works flawless in a JRiver convolution setup with 6 channels. My planned setup with 8 channels (3-way active loudspeaker plus subwoofer) is similar. Additionally should MKV videos be played and JRiver controls video/audio delay. The audio data goes via USB to a RME Fireface UC, in which RME recommend a Dual Core with 2.0 GHz.

Is my actual CPU Intel Dual Core E7200 with 3.06 GHz sufficient for simultaneous audio and video?

[hulkss]

Is my actual CPU Intel Dual Core E7200 with 3.06 GHz sufficient for simultaneous audio and video?

This question has no simple answer. I have my system working very well now with an i7 2600K processor, integrated HD3000 graphics, USB outboard ASIO DAC. I have no video card and no audio card, the i7 does it all. A person could build a tiny HTPC if desired.

When I first started with my HTPC adventure in December I had audio skipping, dropouts, jerky video, terrible lip sync, etc. Now, thanks to continuous development by JRiver and some tweaking, I have performance better than all the gear I replaced. This did require many setting adjustments in BIOS, Windows 7, Intel drivers, ASIO Drivers, VST plugins, FIR filters, and JRiver MC.
I still have a few issues to fix or improve and tweaking with DRC software has no end to it, you just stop and call it good at some point.

[jarbe]

I can point out that the Intel Atom N270 actually is adequate (barely, CPU usage in the 90% range) for running DVD video (mkv files). I can do this if I use convolverVST. But the native convolver pushes the CPU beyond the limit in this case

[stealth82]

I'm not familiar with .PCM files.  Do they have a header?

A 64-bit or 32-bit WAV (which is PCM data with a header) would work if you can convert.

So, do pcm files need to be converted into wav or are they accepted now?
I couldn't understand from the rest of the conversation if you added support for that.

[BerntR]

Matt & JRiver colleges,

Regarding lip sync.

One idea is to include some delay figures in the xml format you have been talking about.

Nevertheless, there will always be unknowns with regards to latency. Distance to the speakers, video latency in the computer as well as in the screen or projector, audio buffers in the computer and sometimes in hardware downstream.

The easiest way to obtain close to perfect latency would be a user friendly graphical interface, where you reduce latency gradually until you just hear that the audio is slighty ahead, then go in the other direction until you just hear that it is slightly behind, and then set the latency in the middle.

This would be a good solution on it's own. A refined version could be be include this as part of the setup. If such a function was combined with latency info in the filter file, JRiver could perhaps adjust the latency automatically whenever a new set of filters were loaded.

Just thinking out loud here.

[Matt]

Nevertheless, there will always be unknowns with regards to latency. Distance to the speakers, video latency in the computer as well as in the screen or projector, audio buffers in the computer and sometimes in hardware downstream.

Remember that lip-sync works nicely with convolution disabled.  This is because all DSP and output components report their latency.  Sometimes hardware requires adjusting manually (for example, my projector has some input lag), which is possible in Options > Video > Advanced.

As for convolution latency, I was wondering if it would work to put impulses through the filters, and calculate how long it takes from input to output.  Then, the convolution engine could just add this time to the latency.  Of course it would use the shortest latency of any channel so that timing between speakers would be preserved.

[Matt]

So, do pcm files need to be converted into wav or are they accepted now?
I couldn't understand from the rest of the conversation if you added support for that.

Only files that can be played by Media Center can be used as filters.  WAV, APE, and FLAC are all good choices for filters.

I don't think we'll try to support RAW files, because it's too messy trying to figure out the bit-depth, samplerate, and channels.  And if we guess wrong, the results can sound terrible (like speaker breaking terrible).

[AudioVero]

As for convolution latency, I was wondering if it would work to put impulses through the filters, and calculate how long it takes from input to output.

This proposal does not include the buffer time of ASIO buffers (or soundcard buffers) and the sound travel time to the listeners seat. With an agreed filter format it would be possible to include the filter delay by an appropriate tag information.

[Matt]

This proposal does not include the buffer time of ASIO buffers (or soundcard buffers)

This is handled automatically by our playback engine.  It is independent of Convolution.

and the sound travel time to the listeners seat

All that really matters is the relative difference between speakers, which would be preserved.  I suppose if the speakers were a few hundred feet away the difference between the speed of light and sound would matter, but in those cases it's reasonable to use the existing lip-sync adjustment in Options > Video > Advanced.

[AudioVero]

As for convolution latency, I was wondering if it would work to put impulses through the filters, and calculate how long it takes from input to output.  Then, the convolution engine could just add this time to the latency.  Of course it would use the shortest latency of any channel so that timing between speakers would be preserved.

Matt,
I've forgotten a point to your statement. If you send an impulse like a Dirac pulse through a filter then you get the filter itself as response. So you can simply take the filter and check for the index of the max. peak to estimate the time delay. As the filter may also switch the polarity (in case of a wrong polarity of playback system or of measurement system) you also have to keep the min. peak in mind too.

[jdubs]

Guys, so I understand the process and its implications more clearly, if I have a 192khz generated filter loaded in MC and I play back a 44.1khz flac, the filter is re-sampled in real-time so that it can be used with the 44.1khz file?  I assume this is why my CPU load goes from 2-3% when using a 44.1khz filter to 10-12% when using a 192khz filter - both playing back the same 44.1khz flac.

Assuming I am understanding what is going on here, correctly, I would definitely vote for filter switching to be implemented (if possible) based on different sample rate material.  Just from a pure CPU load perspective.

Thanks,
Jim

[koldby]

Hi
Me Total Newbee
I just downloaded MC17 to try it out with filters made by Audiolense.
Got it working, but I get some clicks and pops when using the Convolver.
Using external usb soundcard, external usb to I2S converter or internal soundcard makes no difference.
Buffer length is set to max and pre buffer to 10sec.
Nothing wrong without the convolver and nothing wrong when using Foobar2000 and convolver with the same filter.

I really want MC17 to work as I love the feel of it, but for now I stick to Foobar 
Koldby

[koldby]

Well..........
I guess I found the solution.
And somebody mentioned that somewhere in this forum.
Download ConvolverVST and install it.
Problem gone
Koldby

[Trumpetguy]

Well..........
I guess I found the solution.
And somebody mentioned that somewhere in this forum.
Download ConvolverVST and install it.
Problem gone
Koldby

As reported at least by myself in another post - pops and clicks with internal convolver is more than likely a cpu capacity problem. I believe it has been proven beyond doubt that the implementation itself is correct. ConvolverVST is much less cpu demanding. I would try once more and with a shorter filter. The niceties in the 64bit JRiver convolver with a shorter filter by far outweights the old 32bit VST plugin. Especially if you need to change filter for different sources.

[Matt]

In a coming build:
NEW: Convolution automatically adjusts for filter latency for cases where a delay is baked into the filter files (still preserves intentional timing differences between channels).

I don't really understand the point of baking leading silence into all the filters (so it's not for relative speaker distances).  The change above will correct the latency from this.  

We might also be able to skip the convolution step for this leading silence (or even trim the filters of the leading silence to remove the latency) for a performance gain, but really it seems like whatever if building the filters shouldn't be adding this delay.

[jdubs]

In a coming build:
NEW: Convolution automatically adjusts for filter latency for cases where a delay is baked into the filter files (still preserves intentional timing differences between channels).

I don't really understand the point of baking leading silence into all the filters (so it's not for relative speaker distances).  The change above will correct the latency from this.  

We might also be able to skip the convolution step for this leading silence (or even trim the filters of the leading silence to remove the latency) for a performance gain, but really it seems like whatever if building the filters shouldn't be adding this delay.

Matt, for a future build, are you contemplating automatic filter switching based on the playback material's sample rate?

Thanks,
Jim

[Matt]

Matt, for a future build, are you contemplating automatic filter switching based on the playback material's sample rate?

Yes, but I can't say when it might be added.

[jdubs]

Yes, but I can't say when it might be added.

Cool, thanks!! 

-Jim

[Paulv]

In a coming build:
NEW: Convolution automatically adjusts for filter latency for cases where a delay is baked into the filter files (still preserves intentional timing differences between channels).

I don't really understand the point of baking leading silence into all the filters (so it's not for relative speaker distances).  The change above will correct the latency from this.  

We might also be able to skip the convolution step for this leading silence (or even trim the filters of the leading silence to remove the latency) for a performance gain, but really it seems like whatever if building the filters shouldn't be adding this delay.

hello, I'm not an expert but I think leading silences might have to do with linear phase filters and filters with phase correction, minimum phase filter having no leading silences... If I am right I would prefer no change as I am doing phase correction, and linear phase Xover

explanations welcomed
thanks
Paul

[BradC]

Matt,

there is a very good reason for the 'baked in' dalay. The linear phase filters are non-causal (acausal) and apply correction before the music is heard.
You can consider that they work at negative time. Since we can't have negative time, we need a delay.

Uli may be able to explain better.

[Matt]

hello, I'm not an expert but I think leading silences might have to do with linear phase filters and filters with phase correction, minimum phase filter having no leading silences... If I am right I would prefer no change as I am doing phase correction, and linear phase Xover

Timing / phase between channels makes sense to me and is preserved.

But if every filter just starts with 0.5 seconds of silence, that silence does nothing but add latency.

Maybe there's some really low level stuff happening in that 0.5 seconds (pre-echo or something) that's relevant?

[TheLion]

Maybe there's some really low level stuff happening in that 0.5 seconds (pre-echo or something) that's relevant?

Yes indeed, this is the case with linear phase filters. Please don't touch the filter as is and keep bit perfect processing without changes. Every bit of "baked-in" delay is for a very good reason (at least with filters from Audiolense, Acourate and DRC).

Does your latest build change the filter in any way (e.g. cut the leading "silence"/delay) or is it just compensating for any "baked-in" delay? Latter is a great feature. So this does mean I don't have to manually determine and input video delay to compensate for convolution and lip-sync will automatically be spot on (other than hardware delays down the playback chain)?! Thank you!

[Paulv]

+1
please Matt do not touch the filters
linear phase filters are symetrical ie they have as many leading and trailing zeroes

[Matt]

+1
please Matt do not touch the filters
linear phase filters are symetrical ie they have as many leading and trailing zeroes

Leading and trailing zeros have no effect other than a delay.  The math of convolution is just multiplication and addition, and multiplying a number by zero is always zero.

However, I think it's likely that the leading and trailing numbers are near zero but not zero.  For example, in the filter I was testing with, the first sample value is 1.9054464682556382e-016 which is very near zero, but not zero.  We will make no changes to the output in this case.

17.0.83 (available in a few minutes) will adjust the lip-sync automatically for these filters.

[AudioVero]

Matt,

usually filter kernels are windowed to avoid a data step at the start end end of the kernel. Thus the values are quite small. But they are not zero. As stated in the pretty good book at www.dspguide.com we shall not underestimate the influence of small numbers.

[Matt]

I wanted to take a minute and thank everyone involved in this thread.

It's really fun for me when smart users work together with us to build something neat.

In this case we even got generous help from some of the leading experts in the world on audio convolution.

Thank you all.

[hulkss]

Nice work Matt!  

Thanks for making my HTPC project possible.

Brad

[fooze]

Hi everyone,

I've been using MC for a while now, but this thread has finally made me sign up to the forums.

I am very impressed by the users and developers working together to achieve such a great outcome! What an excellent community you have here!

I am very excited about getting my mic calibrated and buying Uli's software so I can at last get a taste of DRC in an easy to use package. Uli and Matt, have you considered partnering with a mic calibrator and writing some fancy scripts to make the world's leading DRC package even more user friendly? I would seriously consider spending $750 - $1000 US on a product that had everything I needed and guided me through setup from start to finish.

Here's to the future!

Matt.

[BradC]

DVD playback:

Is there a solution to get dvds to play without large stutter when the convoled filters have a large delay?

I know that converting all dvds to mkv is one solution, but notmy preferred one.

Ideally there would be a custom video playback setting that would fix the problem

[Matt]

DVD playback:

Is there a solution to get dvds to play without large stutter when the convoled filters have a large delay?

I know that converting all dvds to mkv is one solution, but notmy preferred one.

Ideally there would be a custom video playback setting that would fix the problem

The issue is that the Microsoft DVD Navigator (the thing that reads DVDs on Windows) will not provide the audio more than a little ahead of the video.  This doesn't work well if there's a large audio latency.  People that set the primary buffer size in Options > Audio to a large size run into this same problem.

One solution would be to do DVD title play, which plays the raw MPEG of the main title.  This is what we do when streaming a DVD to a DLNA box.  This solution could work locally as well, but would disable trailers, menus, etc.

Another solution would be to find or write another DVD navigator.  However, it doesn't seem like anyone has made much progress on this, possibly because of the DRM that can be baked into DVD.

[v_erich]

Hi,

when is it possible to use active filters in MC17?
I have a 2-Way with extra Sub with filters created in acourate.
I still have to use an VSTHost for this and cannot use directly MC17.

Thanks,
Erich

[Matt]

when is it possible to use active filters in MC17?

I'm not familiar with "active filters."

Could you explain what you mean in a little more detail?

Thanks.

[AudioVero]

when is it possible to use active filters in MC17?
I have a 2-Way with extra Sub with filters created in acourate.
I still have to use an VSTHost for this and cannot use directly MC17.

Erich,

you can immdiately use the "active filters". 
Simply create a Multiway filter WAV by Acourate from File menu, selecting e.g. Cor1L44.dbl and save it as e.g. ErichCor.wav

Then use a config file with MC like this:

Code: [Select]

44100 2 4 0
0 0
0 0 0 0
C:\AcourateProjects\Erich\ErichCor.wav
0
0.0
0.0
C:\AcourateProjects\Erich\ErichCor.wav
1
1.0
1.0
C:\AcourateProjects\Erich\ErichCor.wav
2
0.0
2.0
C:\AcourateProjects\Erich\ErichCor.wav
1
1.0
3.0

channel 0 is Sub left, hannel 1 is Sub right, channel 2 is Main speaker lft, channel 3 is Main speaker right.