Seed - Guitar Multi Effect Pedal with Neural Networks

Wanted to share my latest guitar pedal iteration using Terrarium and Daisy Seed, which I’m simply calling “Seed”. It emulates 4 amp/pedals using neural networks, as well as reverb, delay, tremolo, and a looper. The enclosure was done using Tayda’s drill, powder coat, and UV printing service. Enjoy!

7 Likes

Very nice demo…Bravo !
Please can you tell me how do you insert the video in the forum…Thank you

Just make sure there is nothing else on the line of you url link, no other character including space or point.

1 Like

The combination of the models with the space and modulation effects sound so great!! The looper is a great new addition too!
And I’m really digging the design too! Looks sleek~

Really enjoyed that metal demo at the end! Passes the heavy test :metal:
(and you started tapping right as I was hoping for some shredding)

Thank you for always keep us posted about what you’re working on! I was so excited to see this new iteration.

1 Like

Nice work! Do you mind sharing some tutorial on writing neural network effects if there any? Couldn’t really find any resources on my side. Thx in advance.

1 Like

Ah this sounds very good! Well done!
Will you be making the .bin available so I can try it on my Terrarium?

2 Likes

@tian here’s an article series I wrote a little while ago on the subject, it’s specific to guitar effects but might give you a place to start: Neural Networks for Real-Time Audio: Introduction | by Keith Bloemer | Nerd For Tech | Medium

@Manysounds I’m not planning on releasing this one at the moment, if I decide to in the future I’ll post here on the forum though.

@keyth72 , thanks for taking the time to share all that and write the articles. The demo is quite impressive.

I read your article about profiling the TS9. One question that comes to mind is if this approach is practical when profiling amps having a realistic number of controls. Let’s consider Marshall JVM, for example. There are six parameters we would like to capture (gain, low/mid/high, and presence/resonance). It would require recording 15625 2 min samples using the same methodology described in the TS9 article. That would take 21 days just for recording the samples alone :dizzy_face:

1 Like

Sorry I just now saw this, enjoying your modules for the bkshepherd pedal! Yes, for modeling knobs, the training data and training time grows exponentially with each additional knob, so you have to be smart about what you try to tackle. Lately I’ll just do a single snapshot model and vary the input gain and that’s good enough for my purposes. Modeling the gain knob with 5 steps is what I do next. I’ve added up to 4 knobs before, a gain and 3 EQ, and I was able to get away with reducing the length of the wav files to about 1 minute each, and only doing 3 steps for each EQ knob. There’s a point of diminishing returns with the sound quality I’ve found. Of course you can alway just throw more processing power at it, there are some plugins out there that have done this for a full 5 or 6 knobs I believe, they just committed to the time for doing all the recording and training!

1 Like

Happy to say that this Seed pedal is now 100% open source! (Except the terrarium pcb I guess), here is the public repository:

There are files for the UV print and a link to the drill template for ordering the enclosure from Tayda. The .bin is available in the releases, enjoy!

2 Likes

I just tried this on my Terrarium - fun!

1 Like