I did a quick test right now: after loading GuitarMidi into Carla, the CPU of Carla spikes to 54% and I get a lot of xruns. If I turn off GuitarMidi (leaving the rest of the setup), Carla CPU drops to 5%.
Do you experience the same issue?
I did set jack to 256 samples.
My jack settings:
samplerate 48000,
Frames/period 256
Periods/buffer 2
Realtime is ticked.
Kernel: 5.4.0-122-lowlatency
I get 22 xruns in 2min 
the cpu usage is quite high, at 54%
The pitchdetection is however doing well enough that its enjoyable to play.
I’m playing with a plectrum simple chord on the higher strings
Chords on the lower strings have problems.
Single notes work well on all strings
Guitar: Les paul with bridge humbucker activated. Gain on Komplete 6 USB interface at 50 %
Carla with GuitarMidi-LV2 connected to Helm
ok, thanks for posting your settings.
I do have basically the same settings and getting the same result.
In my case the xruns affects the sound of Helm and it does not sound nice.
I think that 50% of cpu for GuitarMidi is a lot. As a comparison I used GxTuner and it uses only a tiny bit of cpu.
Do you think that something in the code could be changed to improve the performance?
For sure! The code is not optimized as I am still figuring out the solution. Can you try running the low latency kernel? That makes a difference for me
cool!
I’m already using a low latency kernel (5.13.0-30-lowlatency) and have all other settings as you do.
There is room for optimisation but I doubt that CPU usage will go below 20%. Guitarmidi deploys a filter bank with 148 filters to separate the frequency spectrum. In the next days I’ll push an update which exposes important parameters to tweak so that users can find the best settings for their setup.
Please try the low latency kernel
Can you try the 5.4 low latency kernel? I tried the 5.15 low latency kernel and it was horrible.
Not to blame the kernel though, the 5.4 seems to handle my unoptimized code better than the 5.15 kernel
I tried the 5.4 low latency kernel but I did not experience any improvement.
Can you kindly list all jack parameters? Which synth are you using? Which patch? What are you’re hardware specs (CPU, ram)? Which distro? Do you run jack with realtime priorities?
I’ll try and recreate the problem.
sorry for the late answer, I was quite busy in the last few days.
Here are all my setup invormation:
samplerate 48000,
Frames/period 256
Periods/buffer 3
Realtime is ticked.
Kernel: both 5.15.0-43-lowlatency and 5.4.0-122-lowlatency (no difference between them
the cpu usage is quite high, is over 50% sometimes even 62%
The xruns are now under control. the problem was related to the fact I was using Helm through Reaper, for some reason it was making it unusable. If I use only Helm (any patch) with GuitarMidi, there are no clicks in the audio.
I did run in two other problems though:
Cool that you got it to work.
This is a very hard problem: a note and its higher octave have a lot of the same frequencies. That’s why guitarmidi often interpreted one note as if the higher octave is also played
I hadn’t seen this. I’ll look into it
When you play chords, does guitarmidi work as expected except the two mentioned points?
Can you open issues on github for what you pointed out? Helps me keep track of the bugs
hi,
I opened two issues on github about these problems.
The chords seem fine, I did not notice more problems than what I already reported.
Another issue that I wanted to mention (though I guess you know already): in order to be correctly recognized the played notes must be in the same volume (high volume). This kills the dynamic of playing guitar.
thanks for your work!
I run into this project (for other reasons), but I was wondering if you could integrate such technology into your plugin?
The pitch detection seems very reliable.
Looks really interesting. I’ve decided to improve guitarmidi with deep learning models, so I’ll check this out. I also found this project : guitarml.com and contacted the developer. He hasn’t responded yet though
It is. I think Main prob is that guitarml is potentially CPU heavy due to the neural nets. They would also have to port the vst plugins to lv2 format
No need to port, they already enable LV2 builds using lv2-porting-project.
(when they move to JUCE7 they can use the native JUCE LV2 builds).
Sorry if there was an answer to this already, but is there a way to install this on my mod duo x in the current iteration?
Im sorry. Currently no since it is still in heavy development. I cant really say when it will be available but i am making steady progress