Using convolver universally in HTPC

[antani]

I don't think is a good idea to arbitrarily cut a linear phase filter to reduce preringing. What you cut is definitively not "only silence".
You'll just get a bad filter, latency is required to keep the phase linear.

If you want to reduce the latency, select the mixed phase option in DRC.

[tarsonis]

antani wrote:

It should look similar to this:


Each wav mono filter should have a slot, each channel should refer to a different slot. If you are using stereo wav filters you will use only 4 slots.

Dry is not important, as you said is a bypass, and should be muted. You should only take care of the wet. Each channel's wet level should be adjusted manually. You might want to try A-gain.

But you told yourself it should look like this.
And if I don't cut it, it looks like that

I'm confused what should null samples by -90db do to my room?

[antani]

I did not mean the shape of the filter, just the configuration of Pristine Space.

A linear phase filter needs to be simmetrical around the impulse. The spike has to be at the center. If you cut the filter you loose the linear phase. Try to zoom the filter with Cooledit/Audition and you will see that the samples are not null at all.

Actually even DRC is "cutting" the preringing to create mixed phase filters. But it's much better if you let DRC doing it automatically.

[antani]

Maybe it's worthwhile to give some explanation about linear, minimum and mixed phase FIR filter, without going into math (I could not do it anyway).

When you work in the frequency domain, for example with any equalizer, you also impact the time domain, because the 2 domains are linked by the Fourier transform. If you modify the amplitude of a certain frequency, you’ll also modify its phase.

If you don’t care about phase, but only about amplitude, you can utilize a minimum phase filter. In a minimum phase filters the spike is the first sample of the filter (at the beginning, on the left). You can think about the spike as “music”, what’s on the right of the spike is “after the music”, what’s on the left is “before the music”.

A minimum phase filter can correct only the minimum phase component of your loudspeaker/room transfer function. The minimum phase component is the part that is linear, causal and time invariant. A bass resonance is typically minimum phase. A minimum phase filter does not have any latency because it start to work only from the time “the music plays”.

However a minimum phase filter can create phase distortion. If you want to keep the phase constant you need to use a linear phase filter. A linear phase filter has a symmetric impulse response (the spike is at the center of the filter). It means that it starts to correct before the music plays. Intuitively you might think it is delaying all the frequencies to bring them back in phase. No delay, no linear phase.

A mixed phase filter is a sort of compromise. It can correct the minimum phase component of your transfer function and part of the excess phase. In reality you don’t want to correct everything because the correction becomes too dependent on the listening position. It has some preringing (some samples on the left of the spike), but less compared to a linear phase filter.

Well, more or less .

[tarsonis]

Well your explanation worth gold. Now I understand why this gap is important. I thought a bit off my reverbs for music mixing, where the impulse should be on the start when early reflection is applied. Kind a hard to understand this topic drc. Mainly because drc is not really known in my country and no explaination in my language.
I also think I have made my linear phase filter wrong. I only changed PSFilterType to L and didn't change DLType.

Well, my room ist passive treaded to the maximum. Also my early reflections hasn't really changed with drc. I ain't got problems with reverberation time too. (RT60 < 200hz=472ms | > 200=~258ms) My two main problems are:
combfilter effect 150hz - 1100hz and a dip on left@ 150hz (1 Q -5db max) - dip on the right@83hz (2 Q -7db)
So I tought I could deactivate parameters, like ringing trancation, and set to minimum phase.
DLType = M
PSFilterType = M

If I use minimal latency there is still a gap by in the correction file. Will this gap be compensated by the host/Vst?

I hope I still find the correct settings. I really love to hold the typical genelec sound and I have panik that drc/myself overcorrects my acoustics. Otherwords I don't want to feel myself pressed in a big headphone.

Is there a way to test the filter correction for its correction itselfs, bevor I risk to blow up my speakers?!


//EDIT: Is MPFilterLen equal PSFilterLen and PSOutWindow?
How nessasery is it to measure my main ls to 21Khz, when the filter is only applied up to 20Khz? I never measured above.

[antani]

tarsonis wrote:

... I only changed PSFilterType to L and didn't change DLType. ...

DLType normally is not needed. However linear phase filters are in general an overkill, unless you are introducing strange equalization (i.e brickwall filters). It's better if you use mixed phase. You will get less latency and almost same results.

tarsonis wrote:

... So I tought I could deactivate parameters, like ringing trancation, and set to minimum phase.
DLType = M
PSFilterType = M...

If you want to apply only the minimum phase correction, there is no need to change the type of filter. You can directly take the result of the minimum phase stage with MSOutFile='filename'.pcm

tarsonis wrote:

... If I use minimal latency there is still a gap by in the correction file. Will this gap be compensated by the host/Vst?...

Why are you so worried about the latency compensation? What is the difference if you introduce a latency of few milliseconds? I guess your correction filter is the last thing within your project before audio output, so even in mixing it should not impact. By the way there is no real "latency compensation". No vst host can go back in time. They just reallign different audio streams, but if the filter is at the end, this is not relevant.

tarsonis wrote:

... How nessasery is it to measure my main ls to 21Khz, when the filter is only applied up to 20Khz? I never measured above. ...

Not really necessary. However your Genelec can handle 21 or 22 KHz without problems.

[tarsonis]

antani wrote:

If you want to apply only the minimum phase correction, there is no need to change the type of filter. You can directly take the result of the minimum phase stage with MSOutFile='filename'.pcm

But then no dip and peak limiting is applied? Don't I need them to minimize the combflter effect?!

antani wrote:

DLType normally is not needed. However linear phase filters are in general an overkill, unless you are introducing strange equalization (i.e brickwall filters). It's better if you use mixed phase. You will get less latency and almost same results.

To use mixed phase what do I have to change?
MPHDRecover = Y
MPEPPreserve = Y
Nothing to find in the manuel for mixed phase. Only excess phase. M to D ? I'm getting it wrong again, I think...

antani wrote:

Why are you so worried about the latency compensation? What is the difference if you introduce a latency of few milliseconds? I guess your correction filter is the last thing within your project before audio output, so even in mixing it should not impact. By the way there is no real "latency compensation". No vst host can go back in time. They just reallign different audio streams, but if the filter is at the end, this is not relevant.

Musicians paranoia

antani wrote:

Not really necessary. However your Genelec can handle 21 or 22 KHz without problems.

okay But I don't go down to 10. As I made it, my speaker really made a strange noise to the beginning. I know there is a build-in limiter, but....

[antani]

MSOutFile should include also dip and peak limiting stages. It is just the minimum phase part of the filter. I'm not sure if it includes the mic compensation stage, but I doubt.

If you want a mixed phase filter I think is a good idea to try strong.drc. This is what I am using with audio/video, short preringing but good frequency correction.

[tarsonis]

So, now I know. You have to remove/add the "#" before "XXOutFile".
The mixed phase works very well, but I made a better filter. The only thing is I can't rebuild it anymore,
although I use the same files. o_O
I wanted to change the target. Maybe I have used a parameter that I forget to change. The resulting filter now isn't really boomy as that first one.

I also got the IK ARC System, so I could compare both of that systems.
I really have to say: "DRC rockz!!" The measure results are amazing, and look a lot better as the ARC ones. Well, the sound is much greater than the results look like.
I have 2 filter made with drc that sound 1:1 like the arc filter result, with a bit better center imagination. Awesome.
And DRC is a way faster to measure. Only one point makes max. 6min of measuring.
I don't want to repeat the 3 x 20 measurement at different points anymore. (3 times cause the first 2 measurements sounded very unbalanced)
But I have to say, apart from the prize, ARC comes to good results too. You get a full "don't worry about nothing" package and you don't have learn/understand how "digital room correction" works basicly.
But this could also be a reason to distrust the hole room correction "thing", because off lack of knowledge.

The best results you will get with passive and active room threatment for use in video/music editing. For the home use I would say DRC is a good way to go.

My hope for the future is, that drc gets a GUI or an interactive help manual, which more explains the parameters on the general needs for non-coders. The copy and paste text I read is really hard to understand and optically confusing.
For sure this would be succesfull.

thanks for the help antani and thanks to the people who produced drc

[mchabana]

antani wrote:

In the meantime I changed my Audiotrack with a EMU 1616M, a really nice piece of hardware, with superb performance and simple to configure with Console .

The card seems to be impressive. Does it provide a functionality similar to the directwire i have with my Prodigy card ? If so, what is the name of it, i would like to find some information on the internet ...

Thanks,

Al.

[antani]

It's probably one of the finest pci card available at the moment, at a reasonable price. Its RMAA performance are extremely good and the drivers are very stable even with Vista.

It has the possibility to route back internally the audio streams similarly to Directwire. It's part of the Patchmix functionalities. You can find online the user manual.

[mchabana]

Hi Antani,

i think i will follow your recommandation and go on for the EMU 1616 but since it is quite expensive, i would like to make sure i will be able to use it in conjunction with PowerDVD or ArcSoft TMT; could you confirm that to me ?

I see you are talking about Vista but i think console is not compatible with it; do you know of any alternative to console in that case ?

Thanks again for your time,

Al.

[antani]

I am currently using the EMU with Powerdvd, TMT and Windvd. No problem at all.
Console doesn't work very well with Vista, but still it works. The alternative might be Reaper, a very nice application. I am using it with Mac OS, but I tested it in Vista as well.

[mojave]

J. River Media Center (JRMC) just provided native VST plugin support. I downloaded the Reaper parametric equalizer (ReaEQ) and installed it. I can now provide equalization for the bass frequencies rather than use my Behringer DCX2496. I use JRMC to playback all my media so this seems like a good solution.

However, I am not sure how to use REW to measure the final results. I would need to route the REW frequency sweeps through JRMC and then out the soundcard to the subs and amp. Is there something similar to DirectWire that would let me do this?

[Fujak]

Hello Antani and the others of this thread,

I'm very happy, that I have found this very interesting thread, since I have made a lot of (satisfying) experiences from version 2.6.2 to the 3.0.1 version.
I'm using a Stereo Setup with Isophon Europa Speakers supported by a Subwoofer (Teufel M12000).

I have some question in usage of DRC:

1. I made the experience, that 3 measures with different input level of Mic-Signal in my soundcard produces 3 different results in the final impulseresponse file. How can I be sure to find the exact Mic-Input Level (that means without distortion/clipping on the one hand or level too low on the other hand).
Some additional Informatiion of my measuring setup:
HTPC with ASUS Xonar D2 Soundcard (Stereo-Mode)
Behringer ECM8000 Mic with Behringer Mixer (built in Phantom 48V and PreAmp).
Speaker Level is rather loud (never measured it, but my Speakers can tolerate this level better than my wife ;-) ), so I'm sure having a good s/n ratio
Sweep length is 60sec. played and measured with rec_imp.exe

2. Since my Convolver (SIR1-Plugin) only likes Stereo-Wav-files: Does anybody of you know, how I can use SoX to merge the two output files (L + R) of drc - after translating from raw to wav - into a wav-stereo-file. I know there is a parameter -M, but I don't know the exact usage / syntax; e.g. something like this: sox.exe -M impulsefile_L.wav impulsfile_R.wav -c 2 impulsefile_Stereo.wav.

I have made a scriptfile, where I use rec_imp.exe, drc.exe and sox.exe in order to automate the whole measuring-process with just one mouseclick - but there is still this last step missing to be happy with it.

Any information/suggestion to one of these two questions is highly appreciated

Thanks
Fujak

[jlohl]

you can use/copy the script I did for my software Align that also does the whole process (measurement to DRC calculation) in just 5 clicks.

[Fujak]

Wow, that's very kind of you. Thank you very much. Great work, you have done. It seems to be, as you would have found a possibility to get a GUI for the DRC!
I know, what I will have to do this weekend...

I'm very glad that my problem No. 1 now has got a very good solution.

I also had a look to your correction-file (named as "default") via audacity. There I found, that the amplitude is much higher than of my correction-file. That makes clear, that my problem no.1 is really a problem. I never got such a high amplitude than yours. So I guess that something goes wrong with my configuration settings - especially concerning the settings of my soundcard.

Fujak

[antani]

Fujak wrote:

....2. Since my Convolver (SIR1-Plugin) only likes Stereo-Wav-files: Does anybody of you know, how I can use SoX to merge the two output files (L + R) of drc - after translating from raw to wav - into a wav-stereo-file. I know there is a parameter -M, but I don't know the exact usage / syntax; e.g. something like this: sox.exe -M impulsefile_L.wav impulsfile_R.wav -c 2 impulsefile_Stereo.wav....

sox.exe –M -t raw -r 44100 -c 1 -f -4 filterleft.pcm -t raw -r 44100 -c 1 -f -4 filterright.pcm filterstereo.wav

This will convert the filters created with DRC to a stereo wav file. Use last release of SOX and be careful with conversions, many times the problem is there.

Fujak wrote:

....1. I made the experience, that 3 measures with different input level of Mic-Signal in my soundcard produces 3 different results in the final impulseresponse file. How can I be sure to find the exact Mic-Input Level (that means without distortion/clipping on the one hand or level too low on the other hand).....

Measurement is really key. I noticed that for me it's already a good result if I can record 1 good sweep out of 2. I always check the result of the measurement with the spectral view of Adobe Audition or Audacity.
If you use rec_imp, you can use this command to create the measurement of the sweep:

lsconv sweep.pcm impulseresponse.pcm recordedsweep.pcm

sweep.pcm is the initial sweep, you can create it with glsweep, using the same parameters of rec_imp. For example, this command will create a 44.1 KHz, 60 seconds sweep:

glsweep 44100 0.5 10 21000 60 2 0.05 0.005 sweep.pcm inverse.pcm

impulseresponse.pcm is the result of rec_imp

recordedsweep.pcm is what you want to analyze with Audition or Audacity.

This is an example of what a bad sweep will look like (too many harmonic distorsions):



This is what an acceptable sweep will look like (somehow high noise floor):



The cleaner the sweep, the better is.

[Fujak]

Hello Antani,

thanks for your help. I'm sorry, but the sox-command as written shows the following output-message: "sox formats: can't open input file `-M´: no such file or directory". It seems to be, as sox would identify "-M" not as a global option than rather a name of an input file, doesn't it? I'm using the latest version of sox (sox-14.2.0). Do you have an explanation for this "behaviour" of sox?

Now to the other point, the measuring. I will try this, as you told me and then I will give feedback.
Beside this I figured out, that in the cmd-window there are two interesting parameters to be seen after running rec_imp: peak mic-level and peak value. The Mic Level has to be nearly 0.90000. In case of a higher level, the rec_imp interrupts and warns by a message of clipping. If the level ist too low, rec_imp gives a message, that signal is too low and you should increase the level of preamp gain or playback levels.
The interesting thing is the fact, that the peak value become higher only if the playback levels are higher. Just increasing the preamp gain has only effect on the mic-level. Furthermore I figured out, that the peak-value has to be at least nearly -3.0 dB, because otherwise the level of the correction file for the convolver has a too low signal. So that might be a good orientation how high the speaker levels has to be. But I want to make more experiences, before I can say more.

So far for now.
Fujak

Edit few hours later: Sorry, forget the last sentences; it's not true, as I found out in further tests. The mic-level is shown in the cmd-window as it is set in my soundcard. E.g. when I set line-in level of my soundcard to value 95 same value will be shown in the cmd window as "mic-level: 0.95000". The peak-value however is shown depending on the strength of the input signal from my mic/preamp section. But the peak-value reaches not over -9.0000 and therefore I have to open the gain of my mic-amp at maximum before clipping.
I don't know which meaning this peak-value has. I would be glad, if someone could post the peak-value of his tests/measures as shown in cmd window after running rec_imp or can give some information about the meaning of this peak value.

Another aspect that interests me : my correction_file.wav (for feeding the convolver) in audacity show a amplitude not over +/- 0.5, that means just the half of the possible full range of 1.0. So I have to set the gain in the convolver about +10dB. Is this result normal? Please could somebody of you have a look to your correction files how the amplitude is.? That would be helpful for orientation.

Fujak

[antani]

Fujak wrote:

...thanks for your help. I'm sorry, but the sox-command as written shows the following output-message: "sox formats: can't open input file `-M´: no such file or directory". It seems to be, as sox would identify "-M" not as a global option than rather a name of an input file, doesn't it? I'm using the latest version of sox (sox-14.2.0). Do you have an explanation for this "behaviour" of sox?...

I tried with 14.2.0 and is working fine. However I noticed that if you do a "copy and paste", it is possible that the "-" before the "M" is copied incorrectly. Try to write the command using the keyboard, without copying it.

The amplitude you see in the filter is not important. The gain should be set to avoid clipping, and to keep the stereo image balanced. +10 dB doesn't seem reasonable, you will probably end up with clipping of the audio card.

[Fujak]

Now I have made some tests with lsconv using the option "recordedsweep". To control the sound of the recordedsweep I convert it into a wav-file. I realised some noise of interference and distortion, no matter which settings I have made at my soundcard or the preamp gain / Speaker Output. The best results I've got was with a sweep length of 120 sec., which seems to be very long. But it has the best s/n ratio - and of course the best result of sound.

Antani, I also tested the gain of the convolver. At a gain of +12dB I didn't notice any distortion and music is as loud as without using a convolver. The gain in zero/neutral position produces such a low signal that I have to set the gain at my amplifier very high.

Another test was amplifying the stereo convolverfile with audacity, so peaks are nearby 1.0. The result in convolver: I could set its gain to zero/neutral position and music comes out as loud as without unsing the convolver - witout any distortion. The question is: Do I something bad with the stereo convolver file by just amplifying the wave.

To illustrate I post the recordedsweep (120sec.) and the Stereo-File for the convolver (without amplifying).

[Fujak]

Here are the files:



The sweep starts at about 2:08. The wave before sounds like interference noise. I guess it has to to with the conversion from pcm to wav (made by sox).

The stereo convolver file image shows just a zoom-in of the center of impulse.

[antani]

The preringing of the filter is too intense compared to the impulse. Also the fact that you have to increase the gain by +12 dB is not a good sign. There is no problem in normalizing the filter, but actually DRC should do it automatically.

The filter is probably wrong, but you should easily spot the problem by listening to the result. How does it sound?

[Fujak]

antani wrote:

The preringing of the filter is too intense compared to the impulse. Also the fact that you have to increase the gain by +12 dB is not a good sign. There is no problem in normalizing the filter, but actually DRC should do it automatically.

The filter is probably wrong, but you should easily spot the problem by listening to the result. How does it sound?

Thanks for the fast answer. I hope that it is o.k. having some more questions:
Do you think, the result of the recordedsweepfile is. o.k.?
How can I reduce the preringing in the filter? (with drc_light? I used drc_normal)
The prinringing of the filter is the "wave" before the impulse, right?
How can I do it, making the filter right?

In generally it sounds o.k.but not really satisfying ( a bit artificial).

[antani]

Fujak wrote:

...Do you think, the result of the recordedsweepfile is. o.k.?...

Formally yes, but you have most of the energy in the middle of the sweep (300-500 Hz?). Normally the energy is higher below the Shroeder frequency, where the resonant modes are. Maybe is just due to your loudspeakers.

Fujak wrote:

...How can I reduce the preringing in the filter? (with drc_light? I used drc_normal)...

The problem is not the length, is the amplitude. I think is related to the measurement. Personally I had so many problem with rec_imp that I normally use REW to create the impulse response. It's working fine.

Fujak wrote:

...The prinringing of the filter is the "wave" before the impulse, right?...

Right.

Fujak wrote:

...How can I do it, making the filter right?...

Good question .