If you guys make a comfortable way of sharing snapshots of amps and etc this would be a great selling point for the Mod Dwarf. Lots of people would pay twice the price of the Dwarf for something like that in the Dwarf’s footprint.
This kind of technology is said to be superior both to traditional modeling and profiling. Since models can be conditioned, in theory one could model a complete amp. In practice, given for example my experiments with Pace Bonjour DSP Archetype Australian Gt. Hero, I prefer to create snapshots of the existing presets (without cab, noise gate, modulations, reverbs). Including pedals in front of the amp which are part of the preset. I think model conditioning is important when modeling a real amp. A friend of mine builds real tube amps (I guess I’m not the only one here who have such a friend…). I’m planning to split the amp in pre and power amp and model separately each block, preferably trying to isolate tonestack too so that you can implement it with normal bilinear transform. This could be an interesting approach if Guitarix tries to use this tech for the non linear blocks of their sims that they currently implement with pre-calculated curves and see what they obtain. What do you think @brummer? I would also like to rent a professional studio for a day and record as much Dataset I can over existing outboard so that I can train models later…I asked Oxygen Recording Studio in Verzuolo which is pretty amazing and the costs for such a service is 400eur/day. I don’t think I can afford it. Do you think I should start a crowdfunding? Honestly I feel a little anxious about it…however the final product would be some recorded audio material hosted somewhere…no hw product involved…then everyone could access it…no idea honestly
Could you show us some audio samples? What’s the cpu usage of aida dsp?
If you provide me some di recorded guitar tracks I can record them through the model. Guitar need to be recorded dry directly after the pickup output. At the moment we have only training data that I keep sharing here Aida DSP OS Developers - Google Drive. There are ESR measurements and graphic plots of portion of test signals. ESR vary between 0.008 and 0.011 depending on nn depth. Cpu on nano pi neo2 running 1.3GHz per core (cortex-a53) and using mod software with 128 buffer is respectively 28% and 15% which leaves room for other effects too. I add ir cab and delay/reverbs without problems. We could train with cab enabled really depeends on what we want to do. Another thing you can do is compiling this plugin for mod, which shouldn’t be a pain since I’ve put everything important in cmake files
Here some DI guitar tracks found online
Thank you, time to convert them in wav, cut silence (before playing or between different parts) in Reaper and I will pass through the nn plugin. I have two choices: I can render them directly with python utility script part of the training repo or record through my board (Aida DSP OS). I will be back have a nice weekend. Ah, are those parts copyleft since I would like to include material in training Dataset, do I have to add credits to someone in particular?
Don’t know, found on soundcloud and tgp forum, freely accessible.
@madmaxwell any chance you can upload a zip of your lv2? I’m not getting it to compile and I’m pretty keen to try it out.
Sorry no bin for mod yet, @falkTX is there any mod sdk (with toolchain not the web devel stuff)? That would be easier in this case to release for mod, without need for buildroot build.
No, you need to use the toolchain (with buildroot) in order to build compatible plugins.
I used a docker image to do this the other week and this actually worked rather well.
@madmaxwell Do you have an arm64 binary? I have raspberry pi running mod that I can try it on.
Oh I was just feeling guilty since I didn’t buy the original unit and then @micahvdm we’re in the same league?
Please check Pre-Release dir below to access my night build I’m working on
Please check you satisfy deps below
0x0000000000000001 (NEEDED) Shared library: [libdl.so.2] 0x0000000000000001 (NEEDED) Shared library: [libsndfile.so.1] 0x0000000000000001 (NEEDED) Shared library: [libm.so.6] 0x0000000000000001 (NEEDED) Shared library: [libgcc_s.so.1] 0x0000000000000001 (NEEDED) Shared library: [libpthread.so.0] 0x0000000000000001 (NEEDED) Shared library: [libc.so.6] 0x0000000000000001 (NEEDED) Shared library: [ld-linux-aarch64.so.1]
I’ve implemented support for Neural Models in Mod. I can prepare PR to various repos when we’re ready to go. Do you like the names for the new entries?
@redcloud I managed to cut a little demo with my guitar connected to my hardware (Aida DSP OS) and from here directly in the sound card. This is WIP since I am not a pro player and I probably need to spend endless time on it but I hope it will give you an idea of the plugin.
Regarding the sound, in the song I have three sounds: they’re taken from training over Pete Thorne’s Tone King MKII presets, with all fx disabled. So what you’re hearing is my plugin loading three different model plus mod effects to complement the amp’s sound.
- intro: edge of breakup model plus custom IR Fender Vibro Lux 68 via lsp plugin + Rakarrack reverb (Default, just a bit of mix adj)
- rythm: rock rythm model plus custom IR Fender Vibro Lux 68 via lsp plugin + Rakarrack reverb (room1)
- solo: rock solo model plus custom IR Fender Vibro Lux 68 via lsp plugin + Rakarrack reverb (Default, but almost 100% dry)
No fx from DAW (I used Reaper)
Nice playing and great job! Dynamics sounds great even though the tone is a bit harsh. I’d like to try it with no cab simulation and then going into IR Loader Cabsim, I suspect it would be amazing!
Thanks, I will try to improve the demo in the following days. Consider it’s very much the wrong IR, since the model is basically the amp plus pedal from the original preset. Cab plus equalizer are missing and it’s a huge part of the tone as you know. I would like to record IR separately and I’ll probably need to implement an eq plugin with the same frequencies. Or I could model the bundle, I’m evaluating pro/cons. I’m still experimenting with amp types, neural dsp offers a lot of material, I’m running their demo in virtualbox to save time. Training 5 models takes about 10h so I’m running a script over night after I’ve finished recording all the files.
I’m your first beta tester!
Quick update on the devel side. The plugin is now supporting atom messages and save/restore mechanism so that is compatible with preset saving and custom preset creation. It is also compatible with new file manager model library extension, I will submit PRs in the following week. I’ve also shrinked the code and polished it. I will tag it with v0.9 in the next days. Any news with the binaries I’ve provided? Are they working? It is feasible for beta testing?
It’s sounding Very good!
When I try to add the plugin to pedalboard I get
DEBUG:root:[host] received as response <- 'resp -102\x00'
but I suppose it is because I haven’t set the filesystem stuff yet. I’ve created this folder
but it’s not shown into File Manager. I suppose it should be configured somewhere…