Cuki IR generator Open source App

[Cuki79]

Originally Posted by gitarzysta

Oh I am just quoting what's on your own web page I have no way of knowing what's your advanced algorithm vs online version but I wanted to try and compare the results

"4) If you really want to benefit from my most advanced algorithm, you can record your guitar with both an external microphone and its pickup and send us the recordings"

But there is no way (or I'm blind) I can send the files to you because there is no email/submission form etc - this is why I asked here.

I removed that sentence from the website, and changed it for

4) If for some reason you don't suceed making your own IR or you are not satisfied with the result. You can contact me (Cuki79) through the acousticguitarforum and make a personal request.

My best is not an automated algorithm, I need at least an hour (or two) to get the things done right. So I don't do personal request just for comparison unless the guitar is a must have for the free database.

Try first the online generator, then provide constructive criticism. Some problems can be solved, some can not.

I can not make "better" if I don't know what is the problem...

Again "higher quality" is subjective. You can have a high quality made Gibson and a high quality made Martins, or Bourgeois... In the end, you might like one better than the other.

Same thing applies to IR. Depending on the application, the best IR is different.

Sorry
Cuki

[RJVB]

Originally Posted by Cuki79

I don't think so.

Actually, it looks like Audacity can use the LAConvolver AUUnit on Mac. That should at least give the possibility to get a better idea of how the IR filter behaves in different situations. I have no idea of the math behind this but if it's in any (and I do mean any!) way like modelling an arbitrary relationship with a cubic spline fit to a subset there can be rather unexpected deviations once you go beyond (or too much between) the original datapoints

Audacity on Linux can also use plugins of a number of well-known technologies; if the IR is just a convolution kernel as I think there should be a way to load it in to a filter that knows how to apply them. I've just never seen a convolution kernel in a .wav file (but I should be able to call my Matlab/Octave knowledge back up and add an output handler for a different kind of format).

[jonfields45]

I use Voxengo Boogex with Audacity on my Mac to audition my IRs. The IR/bypass mix control is “backwards” compared to an acoustic guitar pedal, but its convenient audition feature speeds eval.

[RJVB]

Originally Posted by jonfields45

I use Voxengo Boogex with Audacity on my Mac to audition my IRs.

Seems you quoted the wrong part of my post

[RJVB]

Well, I just tried it with the venerable Octave 4.2.2 I can run on my old Linux rig, and a 3min or so segment of a mic + PU recording I had of my archtop. Saved the patch at the highest quality output and applied it via LAConvolver, applied it to a few pure electric recordings ... and got a definitely worse-electric tone. I did reverse the channels to have the PU in the left, and the IR also doesn't make pure acoustic recordings sound electric so something else must be wrong

(I do suppose you could do that if you wanted, make an acoustic sound like an electric or make a quality concert classical sound like Chet Atkins, by reversing the inputs?)

[Cuki79]

1) The algorithm is not made to mimic another instrument
2) It is not appropriate with mag pickup

You can not make the mag pickup of an arch top sound like an acoustic guitar with my algorithm

If you want to do that (or be close) buy a Sim1 pedal

https://sim1guitar.com

The sim1 shapes the transient response of your instrument to match another

My algorithm does not do that.

If you don’t want to pay a sim1 pedal, read their patent and program your own plugin/pedal to do it.

I would say it is a 6 months project. Maybe some extra time to make the pedal.

(You don’t need to make a sim1 pedal… just to make the algorithm to capture the transient shaping parameters and program an IR pedal with a transient shaper).. see my older thread to make the IR pedal and you would just need to program the transient shaping function.

here is the patent
https://patentimages.storage.googlea...80122347A1.pdf

[RJVB]

Originally Posted by Cuki79

1)
You can not make the mag pickup of an arch top sound like an acoustic guitar with my algorithm

Hmmm, I admit I only tried because I had that recording laying around and I saw an IR for a Gibson L48 or so, but now I'm curious why your algorithm wouldn't be capable to determine a mapping from the magnetic sound to the acoustic sound? You can hear that the output from the KA floater is *probably* from a magnetic PU rather than from a piezo, but esp. on the higher strings the sound is a lot closer to the acoustic sound than most piezos can give. IOW, I would have expected the mapping to be easier, rather than impossible.

The main difference between a magnetic PU and a piezo PU that I can think of is that the former gives a more lofi signal, so lacks all those higher harmonics (or parasites) which are responsible for the typical piezo quack (if I understood correctly). Is that the reason, that your algorithm is designed only to remove/reduce those higher harmonics?

[Cuki79]

Originally Posted by RJVB

IOW, I would have expected the mapping to be easier, rather than impossible.

The main difference between a magnetic PU and a piezo PU that I can think of is that the former gives a more lofi signal, so lacks all those higher harmonics (or parasites) which are responsible for the typical piezo quack (if I understood correctly). Is that the reason, that your algorithm is designed only to remove/reduce those higher harmonics?

Please read this:
https://till.com/articles/PickupResponse/

The frequency response of a pickup depends on its position AND on the string.

Here is the response of the neck pickup on the low E string, 82 Hz, 25.5-inch scale, pickup located 6.375 inches from the bridge. The first peak is at 165 Hz, the first null is at 330 Hz.

Here is the response of a neck pickup, 6.375 inches from the bridge, high E string 328 Hz, 25.5 inch scale. first peak is at 659 Hz, the first null is at 1319 Hz.

Both curves are different, it means a magnetic pickup should have two different frequency correction for low E and high E strings... Same with all strings.

You can not train one frequency response correction to do that. The algorithm will do its best... but will fail for sure.

My IR algorithm can model the response of the top, the box and even air and some room reflexions... But not compensate the response of a magnetic pickup.

In this case a Match EQ algorithm + a transient shaper is wiser to implement.

my 2 cents,
Cuki

[RJVB]

Originally Posted by Cuki79

Both curves are different

Shifted in the frequency domain (at least theoretically), no?

My IR algorithm can model the response of the top, the box and even air and some room reflexions... But not compensate the response of a magnetic pickup.

Or the response from a multi-piezo system that has 1 sensor under each string?

[Cuki79]

Originally Posted by RJVB

Shifted in the frequency domain (at least theoretically), no?

Yes but all the strings contributions to the electric signal are sperposed... so you can not isolate them afterward.

Originally Posted by RJVB

Or the response from a multi-piezo system that has 1 sensor under each string?

That you could. There is a Roland/BOSS patent about the GR midi devices that describes the hexapickup compensation. But I don't remember which.