Live data from Hacker News

Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

github.com

61–70 of 163 posts

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#61
post #44

For people who don’t play guitar, this is really commonplace technology now. Engineers worked out how to model the circuit behavior of any amp a while ago. Powerful valve guitar amps have really interesting effects on the waveform particularly when driven hard, but that valve technology is expensive, unreliable, bulky, and very loud when producing some of the most sought after tones. The most popular guitar amps now…

> The most popular guitar amps now are all solid state

Maybe on the low end of the market, but tube amps still reign supreme for most serious players.

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#62
post #16
post #8

What a clickbait title. Sounds (to whom?) like a $600 (which one?) amp using deep learning (why does it matter?). To expand on this: -$600 tube amp is not description of a sound. And a plug-in is not something you plug your guitar into to make enough noise for 300 people. "A plug-in sounds like a $600 speaker" makes almost as much sense as the title; by the sound of an amp people often mean the sound of the combo box…

It's based on a Fender Blues Jr according to the vid. It sounds pretty good for a free plugin, but as you said there's tons of free VSTs and cheap solid state amps that can do better. I think it sounds good for a hobby project though!

What free vsts for guitar amp sound better? I'd like to know

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#63
post #60
post #6

Just tried it out, I don't know if I would say it sounds like a $600 tube amp especially in high gain scenarios (it really didn't like my active pickups), for a free plugin it is definitely great. In comparison to the Archetype Nolly plugin from Neural DSP it doesn't really compare, but but for something thats free and can be trained at home on custom audio samples its pretty cool.

wow, I was literally just about to download the trial for the archetype nolly (after being iffy on whether the fortin nameless was right for what I wanted), what are the chances it's being discussed on HN. I'm hoping that the nolly plugin can cover both dirty and clean sounds well enough to not need a second plugin, fingers crossed

The Nolly is awesome, tons of options in there for both clean and dirty sounds. The awesome thing is you can try all their plugins before buying to make sure it’s what you need.

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#64
post #12

This is really cool, I’d love to try this out! Neural DSP, developers of what are arguably the best amp sim plugins available right now, claim that their upcoming floor modeler uses what they describe as “unique biomimetic AI technology” to create its sounds. I wonder how different their work is from this. https://neuraldsp.com/products/quad-cortex I’m personally kind of over lugging an amp around and if live music e…

Thank you! Any feedback is appreciated if you get around to trying it out. If you look at the WaveNetVA repository from Aalto University that my plugin is based on, at least one person affiliated with that project works at Neural DSP. I suspect that the QuadCortex uses something like this (maybe a more advanced neural net) combined with other proprietary techniques and application specific hardware. They claim they c…

It is really interesting to see this becoming more widespread. I am waiting for a delivery in the first batch of the Quad Cortex.

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#65
post #61
post #44

For people who don’t play guitar, this is really commonplace technology now. Engineers worked out how to model the circuit behavior of any amp a while ago. Powerful valve guitar amps have really interesting effects on the waveform particularly when driven hard, but that valve technology is expensive, unreliable, bulky, and very loud when producing some of the most sought after tones. The most popular guitar amps now…

> The most popular guitar amps now are all solid state Maybe on the low end of the market, but tube amps still reign supreme for most serious players.

There are plenty of touring musicians using Kemper digital models of their tube amps or Helix/AxeFX

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#67
post #50

Earlier quoted context omitted.

Nowadays it's better to directly start using an ESP32, if you want to experiment. It has bidirectional I2S, (for example mic in, apply dsp, then out to speaker), is dual core, which would allow you to do the audio-related stuff on one core, and everything else (like communications) on another core. If you then see that you don't require all the good things an ESP32 has to offer, maybe because you only want to drive a…

Why go for a esp32 if you can use i2c for reading many values and driving relays? Serious, because I am building a complex product and trying to keep costs down. Or are you referring to hobbyists?

If you're making something for audio you need I2S to talk to the DAC/ADC.

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#68
post #58
post #15

Would be cool if you could run this thing on a raspberry pi and actually use it like an on stage amp

That's unlikely to end well. You really need custom audio hardware - or at least dedicated commercial-grade stuff, for this sort of application. Latency is HUGELY important. 2-3ms can be felt, and most consumer level stuff uses buffers WAY bigger than that.

As a reference the new kid on the block, NeuralDSP’s Quad Cortex uses 500 Mhz Quad-Core SHARC https://www.analog.com/en/products/processors-microcontrolle...

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#69
post #14
post #9

Earlier quoted context omitted.

IANAG, but it sounds like it's recorded through a speaker in the room, and the miking wasn't optimal. Now that might be the recording we watch, or the actual output of the plugin; no way to tell. But apart from that, it's quite good, IMO. (I Am Not A Guitarist)

As a guitarist, it is important to understand that any time you tell a guitarist that the sound is digital, the guitarist will scoff and say it doesn't sound like real tube amp, regardless of what they said or thought before knowing it was digital.

Small anecdote in the similar vein.

I gigged around Boston in the early 2000's playing guitar in various rock and metal bands, usually in front of 15-20 people. I had been using a 1960's Fender Bassman and changed out the tubes to some Sovtek ones. Sound guy at one of the clubs (who's had been around for 20+ years at that point and seen everything) noticed right away and asked why my amp sounded off. A little while later I bought one of the first Line 6 modeling amps and my first gig with it the same sound guy said "Ahh, you put the old tubes back in". I'm like, "Nope, totally different amp, and it's all digital.". He didn't believe me and while he was inspecting the amp kept saying "No, no, no. There's tubes in here I can smell them warming up".

Re: Show HN: SmartGuitarAmp – Guitar plugin made with deep learning

#70
post #50

Earlier quoted context omitted.

Nowadays it's better to directly start using an ESP32, if you want to experiment. It has bidirectional I2S, (for example mic in, apply dsp, then out to speaker), is dual core, which would allow you to do the audio-related stuff on one core, and everything else (like communications) on another core. If you then see that you don't require all the good things an ESP32 has to offer, maybe because you only want to drive a…

Why go for a esp32 if you can use i2c for reading many values and driving relays? Serious, because I am building a complex product and trying to keep costs down. Or are you referring to hobbyists?

The context of this thread is audio processing, so having I2S (Inter-IC Sound, with DMA) on board is really valuable, as is having an additional core.

In my case, I can order a bundle of three esp8266 13.79€, or a bundle of three esp32 for 17.99€, both as the most lazy, fastest delivery option on Amazon. If you know that you don't need the extra features of the esp32, save yourself the 4.20€ (1.40€ per board) to keep the cost down. But if you're new to it and want to explore the possibilities of a modern µC, it would be a mistake to save in the wrong place.

Even consider getting a LOLIN32 Lite (three for 19.99€), since the width of the board is smaller than the normal esp32 and esp8266 dev boards, which then lets you use both sides of a breadboard to plug in cables. All the other boards are so wide, that on a normal breadboard you will have to decide which side of the board you want to have a row of externally accessible pins, and on the other side all the cables must be plugged in below the board (search for "LOLIN32 Lite width breadboard")

Saving on cost can also be a premature optimization.

> Or are you referring to hobbyists?

Any context that is not production. Development, Education, Hobby.

Post reply on HN