Multifilter MeltDown ?

[Marc Hildebrant]

Group,

I'm processing a stereo recording using a multifilter. In the multifilter are two CNF's. Preview yields a "choppy" playback and when I process the multifilter, the result is a portion of the original signal with "chunks" removed.

The samples to process window shows 800000.

What am I doing wrong ?

Regards,
Marc

[Craig Maier]

Hi,

I just tried two CNFs in the multifilter in version 7.5, and it worked normally both in Preview and Run modes. Were there other effects or filters in the chain when you experienced this?

Craig

[Marc Hildebrant]

Craig,

No, the multifilter consisted of just two CNF units chained together. If I have a mono channel, then the two in series work fine. If I have a single filter on a stereo track, no problem.

If I have two in series on a stereo track, then the preview "sputters" and the processed track only contains "chunks" of the source.

Other multifilters that I have are quite long and do not have any problems.

By the way, I use two CNF units to remove the distortion and remove the noise. Both filters are set to the same values and use the 16K samples.

Regards,
Marc

[Craig Maier]

Hi Marc,

Well, one difference is that I used version 7.5 and I am not sure what version you are using. But, using two CNFs in the multifilter is completely legitimate and often a good idea for the reason that you alluded to. So, I am not sure what could be going on there at this point. I wonder if it is parameter sensitive? Maybe you can provide the settings for me to use in my test.

BTW - - - my test used a stereo wavefile, 44.1 kHz, 16 bit and fast edit mode. Does that match up with your system setup?

Craig

[Marc Hildebrant]

Craig,

The samples are 16 bits at 44.1KHz. I am using classic editing mode. Software is 7.15 DC7

I tried increasing the virtual memory size and some other system tricks to no benifit.

The computer system is about 10 years old...512 M ram and a 1.3 GHz CPU.

Regards,
Marc

[Craig Maier]

Hi Marc,

What is the operating system?

Craig

[Marc Hildebrant]

Craig,

Windows XP..."Professional"

Regards,
Marc

[Craig Maier]

Hi -

I just tried running 2 CNFs in the Multifilter, only this time in classic edit mode, but again using version 7.5 (and I also tried it on DC8 Beta 4). I do not have your version of DC8 on this machine anymore because we had been using that as a development platform for DC8, so the .exe is not the same as yours. I am running the XP operating system too.

I would think think that your CPU is fast enough for this task, and even if it was not, the system would only loose chunks during preview and not during a run command.

So, I do not know - - - but I will grasp at a straw - - - the RAM that you have on the machine does seem to be pretty small. This machine is using 2 GByte, and so that is one difference that could be in play - - - really not sure though. Maybe someone else can make comment on that aspect of things.

Are the CNF settings unusual or extreme? For example, are you using the highest number of ffts allowed? Does lowering the fft size (or any of the parameters) make a difference?

Craig

Craig

[Marc Hildebrant]

Some Answers

Craig,

The CNF units are both set to 16384 for FFT size. If I set them to 1024, then the "problem" stops.

Looks like I have a limitation due to my computer.

Regards,
Marc

[Craig Maier]

Hi Marc,

I am not positive that you have a system limitation. I just noted a difference between this machine and yours (besides CPU clock speed). It could still be something else.

Here is something to check - - - does it run ok at 8192 ffts? How about 4096 ffts, etc? At what value does the system run into trouble?

Craig

[Marc Hildebrant]

Yet More Info

Well, it turns out that the ORDER of the CNF devices/samples are important.

Case One:
Two CNF devices in the multifiler. Stereo Recording.

First Filter at 16384 samples/50% overlap. Second filter at ANY sample setting and problems occur.

Case Two:
Two CNF devices in the multifilter. Stereo Recording.

First Filter at 8192 Samples/50 % , Second at 16384 samples. NO problem.

Case Three:
Two CNF devices in the multiufilter. Mono Recording.

Both filters at 16384 samples...No problem.

Case Four:

One CNF devices with many other filters. Stereo Recording.
CNF at 16384 samples. No problem.

Looks like it's related to how the signals go into the frequency domain ?

Regards,
Marc

[Craig Maier]

Question:

What happens if both CNFs are set to 8192 ffts?

Craig

[Marc Hildebrant]

Craig,

If both at 8192 no problem. Case two showed that you could increase the second filter with no problem. It's the first filter in the string that sets up a limit in my system.

Regards,
Marc

[Craig Maier]

Marc,

It could be some sort of buffer assignement related issue (not enough buffer) and/or a bug in the software - - - not really sure.

But, in general, using 16384 ffts is not generally recommended. While large fft counts provide maximal frequency resolution, it also provides the poorest time domain resolution which results in the poorest transient response. Secondarily, it also probably has the highest propensity to generate artifacts. The system was designed to perform optimally at around 2048 to 4096 ffts. Give it a spin with those values and it should sound the best in that region (depending on the material).

Craig

[Marc Hildebrant]

Craig,

O.K. then, I'll try a lower value.
The 16384 number came up when I was working to optimize the noise removal and still preserve the low end. Also, it helped to clean up the "IM" distortion with the artifact surpression on . By the way, the clean up of the "IM" or what I call trumpet distortion is a great tool.

Regards,
Marc