Wish Optimize far field IR processing (if not done already)

[fractalz]

So I was wondering why one of my big presets is taking so much cpu (hovering around 85% on my mkII Turbo) and I remembered that I put some FullRes room IRs in the cab block (4 total cabs).

Looking at the align tab, I can see that they are indeed very long, but they start with a long span of 0s... Got me thinking, maybe there is a way to analyze these files and note where the signal is above some low background noise level and skip processing the leading samples and only convolve the "signal".

Maybe this isn't how things work, or already done, but I thought I'd throw it out there.

The room IRs add 10% each, taking my 65% preset to 85%. Since over half of the Room IR is zero, maybe there's a good amount of savings to be had.

 

[Greg Ferguson]

From what I remember, if you have a single IR in the Cab block, the IR will be trimmed. If you have multiple IRs, that doesn’t happen.

I pull ones that haven’t been trimmed into Cab-Lab, let it trim them, then I save them and import them, though, if I was trimming two or more I’d consider using Cab-Lab to blend them and save the resulting IR and use it instead of the two separate ones because it will take less CPU.

I made a cursory pass over the master wish list that @yek maintains and didn’t see anything requesting the additional of this functionality to the Cab Manager.

 

[fractalz]

From what I remember, if you have a single IR in the Cab block, the IR will be trimmed. If you have multiple IRs, that doesn’t happen.

I pull ones that haven’t been trimmed into Cab-Lab, let it trim them, then I save them and import them, though, if I was trimming two or more I’d consider using Cab-Lab to blend them and save the resulting IR and use it instead of the two separate ones because it will take less CPU.

I made a cursory pass over the master wish list that @yek maintains and didn’t see anything requesting the additional of this functionality to the Cab Manager.

I think that is trimming the tail of the IR.

I'm suggesting "trimming" (really, skipping) the leading samples in a far-field IR.

It may be that this isn't a valid idea depending how the convolutions work...

 

[Greg Ferguson]

I'm suggesting "trimming" (really, skipping) the leading samples in a far-field IR.

I understand that’s what you’re talking about.

When Cliff released his sample IR the other day, this was mentioned …

Interesting.... the first peak is at about 140mm in the Align tab. In order to blend with other IR's I had to bump them to 140mm to stop the phasing. Is this by design? […]

I tested and saw the problem. Opening the IR in Cab-Lab with the trim option on showed no leading gap. I saved it and, IIRC, don’t see the problem now with my copy of the IR. I’ll dig into the presets that I have that are using the IR to compare to the downloaded copy.

Auto Trim – this removes superfluous silence from the start of the IR. Many commercial IRs do not require this. It might come in handy when shooting your own IRs using IR Capture.

"There is no wrong place to trim. It's impossible to know where the data starts because of noise. So we find where the data starts to increase, back up a few samples and trim there." [44]

I think that with a single IR in the Cab block the IR is automatically trimmed and we can’t see it because there’s nothing else to compare it to or align with. If we add a second IR then auto-trimming is disabled so we can align manually. The manual is not clear about it, but that’s what it seems to be doing.

Back to the original issue of two IRs with leading whitespace, it looks like they are raw IRs because they are not trimmed and they are out of phase. What to do when you can’t line up IRs? talks about it. I don’t know if Cab-Lab Lite or Cab Manager in Axe-Edit can do the Minimum Phase Transformation that is needed but Cab-Lab can.

 

[DLC86]

So if an IR has leading space it might have noise preceding the real (desired) sampling and the algorithm is being tricked.

Yup. And I've found various examples with this issue, from several vendors.

 

[FractalAudio]

Those are not "far-field" IRs. Those are room IRs. They are nearly 1.4 seconds of data. It takes a lot of CPU to process 1.4 seconds of data with zero latency. The leading silence is a tiny fraction of the total IR length.

 

[fractalz]

Those are not "far-field" IRs. Those are room IRs.

Thanks for the reply. Sorry for confusing the terminology.

They are nearly 1.4 seconds of data. It takes a lot of CPU to process 1.4 seconds of data with zero latency. The leading silence is a tiny fraction of the total IR length.

I didn't realize the align view didn't show the full IR (didn't really look at the time axis). Agree, trimming the 7ms of "silence" won't help on a 1.4s IR.

Looking at the IRs above, they seem out of phase. Is that desired? These are factory IRs : Legacy 194 & 195.