FREE Custom IRs for Your Guitar and Its Pickup (Acoustic Guitar IR Impulse Response)

[jonfields45]

Originally Posted by Cuki79

Ah Ah[*] Jon's algorithm has a built-in threshold. If the pickup signal is below a certain level, instead of making a risky decision on how to transform the signal, it decides to leave the signal unprocessed. It means that Jon's algorithm has a "built-in blend parameter". That's why it is so dry and has a nice fast response with the cost of mimicing less accurately the mic.

To be specific, what I do is search the pickup data in the frequency domain for its largest component, then I go through the rest of its frequency components and look for ones more than 65 dB down from that peak. When I find these small numbers I eliminate them from the IR (set the IR to 0 dB for that frequency band). Dividing by small numbers in the final IR calculation (mic divided by pickup) seems like a bad idea to me (two small numbers could ratio to something significant and pull the IR off following what is just noise).

The downside is I also make sure the sample audio peak is within 15 dB of clipping. Starting from 96 dB and taking away 15 leaves enough remaining dynamic range for my -65 dB test to make sense and work. I log how many of these "near zeros" were found in IR generation. Given the 24KHz bandwidth of 48KHz sample rate, only maybe 8 KHz of guitar, and 64K frequency bands in my script as written, I like to see 25-40K near zeros reported by my script. When I don't I usually find some sort of error in the recording.

However, miss matched pickup and mic volumes should only impact the IR by giving it the overall gain of that miss match (remember it is attempting to make the pickup sound like the mic, and if louder for example is part of it, then louder it will be).

Inspired by this discussion I made recordings of my Traveler Escape MKIII and attempted direct IR generation. It was shocking to me how closely the IR worked making the pickup sound like my mic even if I used some pretty heavy handed EQ on the mic recording (for example I tried putting in a 100-200 Hz 15 dB peak like a real acoustic guitar). This solid body guitar has a piezo bridge pickup which is typically IR friendly...

However, none of the mic recordings sounded good enough to me and the resulting IRs were very good facsimiles, but not worth the trouble.

[kevinplarson]

I’m now using the LR Baggs Voiceprint which isn’t the same I know, but how important is it to resample and make a new IR with a dramatic change of strings? Like going from Phosphor Bronze to Monel (retros)? It is a pretty different sound.

[Cuki79]

Originally Posted by kevinplarson

I’m now using the LR Baggs Voiceprint which isn’t the same I know, but how important is it to resample and make a new IR with a dramatic change of strings? Like going from Phosphor Bronze to Monel (retros)? It is a pretty different sound.

No impact.

Here are your "two signal paths" on the picture below. The IR will try to compare both and compensate the pickup signal path to mimic the acoustic signal path.

The strings are common to both paths, so the IR has nothing to compensate for them.

[Doug Young]

Originally Posted by kevinplarson

I’m now using the LR Baggs Voiceprint which isn’t the same I know, but how important is it to resample and make a new IR with a dramatic change of strings? Like going from Phosphor Bronze to Monel (retros)? It is a pretty different sound.

Just to make Cuki's point a different way - what these IR products are trying to do is to learn the difference between 2 sounds and then create a transformation that makes one sound like the other. Once it's learned that difference, changes like strings shouldn't matter. Maybe new strings are brighter, for example, but the difference between what you hear acoustically and what is picked up by the pickup is still the same. You can change the sound dramatically by changing your hand position, or using a different pick, or strumming vs fingerpicking - none of that should matter either, because that change affects what you hear acoustically and what the pickup gets.

The only way I could think of the strings mattering is if somehow you have a string that reacts differently with the pickup - so that the pickup sound changes in a way that's a lot different from the change you hear acoustically. That could possibly happen with a magnetic pickup and strings made of different metals, but doesn't seem likely with other types of pickups.

[kevinplarson]

Thanks, guys. Makes sense.

[Lid555]

Thanks again Jon and Cuki for the excellent and detailed descriptions of how your algorithms work regarding volume differences.

In reading over your responses (which I’m still working on trying to fully understand), I realized I was forgetting one thing in my setup that is maybe relevant to IR. I use an active circuit inside my guitar to buffer and amplify my piezo signal. Do buffer and pre-amp circuits play into the IR equation? Or can we basically treat them as being linear? I'm definitely not trying to push the amp circuit to saturation- I'm keeping things as "clean" sounding as possible. But I did notice the piezo signal is much less dynamic than the mic signal, though maybe that's just the "natural" response of the piezos?

Cuki, is there a way to input recorded audio files into your algorithm? I am very much wanting to work with IRs like how Jon has set up his code with Octave (i.e. in the manner that I can edit/process the same audio files and generate new IRs from there without re-recording a performance). Maybe I can fake the live input with Soundflower somehow?

[Cuki79]

Originally Posted by Lid555

I use an active circuit inside my guitar to buffer and amplify my piezo signal. Do buffer and pre-amp circuits play into the IR equation? Or can we basically treat them as being linear? I'm definitely not trying to push the amp circuit to saturation- I'm keeping things as "clean" sounding as possible. But I did notice the piezo signal is much less dynamic than the mic signal, though maybe that's just the "natural" response of the piezos?

By definition a buffer should be fully linear. To my knowledge only the Takamine CoolTube preamp adds distortion. Most other preamp are meant to be linear. However many preamps include a high-pass filter, a low-pass filter and many includes an EQ to colour the tone. The IR will do its best to compensate but it can not amplify what is gone. That's the limit.

Originally Posted by Lid555

Cuki, is there a way to input recorded audio files into your algorithm? I am very much wanting to work with IRs like how Jon has set up his code with Octave (i.e. in the manner that I can edit/process the same audio files and generate new IRs from there without re-recording a performance).

Uncomment lines 231 to 234

File must be stereo
Pickup track must be on the left
Mic track must be on the right

I answered it in the "closed issues" on the github.

Beware, the open-source algorithm is a simplified version of my current algorithm.

[Lid555]

Great, thanks for the info Cuki! My buffer/preamp circuit does not have any additional processing so it shouldn't be an issue.

Thanks for the info on how to work with your algorithm with files. I'll test some stuff out in the coming weeks and report back here. Since my application here is a little non standard (piezo for clean electric guitar) I'm thinking bare bones algorithms might work best anyhow (Jon's and your simplified version).

[-GF-]

@cuki79 and @jonfields45 Have you seen the new Zoom unit?

https://zoomcorp.com/en/gb/multi-eff...four-g2x-four/

.... Sounds interesting:

MULTI-LAYER IR
Each amp model is built with Zoom’s innovative Multi-Layer IR’s, to recreate the tonal characteristics and feel of playing through classic amps and cabinets. These Impulse Responses were captured from iconic cabinets at 3 different volumes and are blended and activated based on the volume of your guitar.

[jonfields45]

Seems like an interesting idea. Might be more marketing than a practical improvement.

[Cuki79]

It probably models the speaker breakup… my assumption is that nobody wants to listen to guitar top breakup…

[laenzi]

Hi @jonfields45

I am software engineer and musician and I wanted to ask you if it would be ok for you if I would implement your open source algorithm in Rust so that it could be easily executed on any modern computer (e.g. from the command line)

I would like to create a library which could also be used in a LV2 plugin. (That would be the next goal)

My motivation is of course the music, but then also to provide such great technology to the community. I also want to refresh my knowledge about DSP and i want to learn the Rust language and LV2 technology.

I think rust is a great language to implement such algorithms.

I suggest to implement it under GPL3 license.

Please feel free to contact me in case of any concerns from your side.

[jonfields45]

Originally Posted by laenzi

Hi @jonfields45

I am software engineer and musician and I wanted to ask you if it would be ok for you if I would implement your open source algorithm in Rust so that it could be easily executed on any modern computer (e.g. from the command line)

I would like to create a library which could also be used in a LV2 plugin. (That would be the next goal)

My motivation is of course the music, but then also to provide such great technology to the community. I also want to refresh my knowledge about DSP and i want to learn the Rust language and LV2 technology.

I think rust is a great language to implement such algorithms.

I suggest to implement it under GPL3 license.

Please feel free to contact me in case of any concerns from your side.

Jim Amsden just translated my script to Python. Having turned 65 last summer, Matlab is sort of my era :~). Here's what he sent me. I'll put it up on Cuki's website (my open source article) in the next day or two (unless one of you guys finds a problem).

https://drive.google.com/file/d/1zT6...usp=share_link

EDIT Also via Github:

https://github.com/jamsden/generateBodyIR

[Buckeye Slim]

Originally Posted by Cuki79

....

Aaron Short compared recently different IR pedals that "can" capture the IR:

Reading thru this thread has been quite educational. Watching this video really helped demonstrate what the deliverable of these efforts is. To that end, I have a few questions.

• How is the Cuki's IR with EQ Match accomplished? Do you live with the original IR output and then EQ it as an effect on top of what's coming out of the IR pedal?

• Is it possible to apply this desired EQ as part of the process of generating the IR so that the resulting IR file contains it rather than having to post-process it?

• If you want what the IR file produces to be different than what it produces from comparing the pickup channel to the mic channel, which input (or both) should you play with to modify its input values? Let's say that you agree that you like the sound of the mic'd version better, but even that version leaves some room for improvement. Do you do something like make the contrast between the pickup and the mic in the area you want to "improve" even greater so the algorithm boosts the adjustment? Or is it preferable to post-process the output from the IR pedal to finish it off?
  
  A ton of valuable information about what's going on when amplifying an acoustical instrument. Thanks!
  

[laenzi]

I have pushed the work on github: https://github.com/laenzlinger/irgen

i have just started with the algorithm of @jonfields45. I hope thats ok for you guys, otherwise, just let me know

The idea is to also implement other algorithms in future.