New Reverb Published by me

B

BasS

Member
Hey everyone, I’m pretty new to doing Teensy projects, but getting really stuck in. I wasn’t super impressed with the stock reverb (designed for a different era with different hardware, not the 600 MHz DSP we have now!) so figuring this would be a well researched field with plenty of papers on how to design a nice reverb, I asked Claude to make me one. And I think it did quite a reasonable job. (I’m being modest here…)

I published it MIT licensed as an Arduino library at https://github.com/basscheffers/VERBtrix

If you want to hear a demo, here it is with the EP patch of an analog synth I am cloning: (dry first, then with reverb. Excuse my playing skills, or lack thereof.)


(Which will also be shared in due course)

Here it is on the prototype of the hardware for my synth:

image0.jpeg
 
For the record: I am professional C++ developer with 30 years of experience.

I have great deal of problem with code entirely written by AI/Claude.

Anyone can prompt AI to generate code and it will come up with something more or less "working" and that is fine for private amateur use,
but the moment that you publish it as your own creation and say you license it under "MIT license" the problem starts.

Neither you, nor Anthropic or whatever other AI vendor, have any copyright on generated code. Therefore you just don't have rights to assign in first place. US Copyright Office guidance (and Thaler v. Perlmutter, which the Supreme Court declined to review in March 2026) is now settled: copyright requires human authorship. Code produced by a machine with no meaningful human creative input isn't copyrightable. Prompting is not enough.

Therefore saying that it is "MIT licensed" is just not true. If this was prompted and shipped largely as-is, there's no copyright to assign, which means the MIT license is essentially meaningless - you can't license rights that don't exist. The closest to truth might be "it has no copyright". Also various jurisdictions might treat this issue in different ways.
 
Last edited:
I too am a professional developer with 30 years experience. And tagging something with MIT “license” is essentially saying “there is no copyright”.

So I’m not exactly sure what “problem” you think there is.

it’s a good sounding reverb, pretty efficient with resources. Anyone can use it. That’s all there is to it.
 
1. MIT license is NOT equal to "no copyright". "No copyright" is equal to public domain. You don't have any rights to AI-generated code, therefore you can't assign the license. Slapping an MIT License on it has no legal weight because there is no copyright to grant permissions for. The code is effectively in the public domain the moment it is created. You can't even put your name on it.

2. The code is not efficient on resources. As in typical AI-slop it is over-engineered. For example using 6th order Lagrange is over the top and waste of resources. AI did not ever hear any sound, it doesn't know what is audible and what is not.
 
Last edited:
Sounds good, there are some excellent libraries that are open source here for the teensy too.

Look forward to see what becomes of the synth.
 
tomas said:
Neither you, nor Anthropic or whatever other AI vendor, have any copyright on generated code. Therefore you just don't have rights to assign in first place. US Copyright Office guidance (and Thaler v. Perlmutter, which the Supreme Court declined to review in March 2026) is now settled: copyright requires human authorship. Code produced by a machine with no meaningful human creative input isn't copyrightable. Prompting is not enough.

Therefore saying that it is "MIT licensed" is just not true. If this was prompted and shipped largely as-is, there's no copyright to assign, which means the MIT license is essentially meaningless - you can't license rights that don't exist. The closest to truth might be "it has no copyright". Also various jurisdictions might treat this issue in different ways.
Click to expand...

First time poster because I feel like I finally have something I know about and useful to share. Software developer (25 years) and lawyer (bar in California) here. I also give expert witness in software copyright cases.

US Copyright guidance and Thaler v. Perlmutter does NOT say AI generated code cannot have copyright. It merely says the sorely AI generated works do not have copyright on their own. If @BasS used human judgement to guide the AI in what to build that almost certainly is copyrightable.

Thaler v. Perlmutter only rules that a human must be listed as the author for it to be copyrightable. To quote:
The United States Court of Appeals for the District of Columbia Circuit reviewed the case and affirmed the district court’s decision. The court held that the Copyright Act requires all eligible works to be authored by a human being. Since Dr. Thaler listed the Creativity Machine, a non-human entity, as the sole author, the application was correctly denied. The court did not address the argument that the Constitution requires human authorship, nor did it consider Dr. Thaler’s claim that he is the author by virtue of creating and using the Creativity Machine, as this argument was waived before the agency.
Click to expand...
See https://fairuse.stanford.edu/case/thaler-v-perlmutter/ (and note this is NOT what @tomas claims it says). There is no attempted copyright registration under Claude's name here, so we can dismiss this part of @tomas argument.

The US Copyright Office Guidelines gives broader guidance:

As explained in the AI Registration Guidance, “a human may select or arrange AI-generated material in a sufficiently creative way that ‘the
resulting work as a whole constitutes an original work of authorship.’” [snip] A relatively common scenario in registration applications is the combination of human-authored text with AI-generated images. In one early case, for instance, the Office found that the selection and arrangement of AI-generated images with human-authored text in a comic book were protectable as a compilation
Click to expand...
(page 24 of https://www.copyright.gov/ai/Copyri...telligence-Part-2-Copyrightability-Report.pdf)

Provided @BasS used human taste, judgement and creativity or directed the architecture to build the body of work they shared the work as a whole is probably copyrightable (I say probably because this isn't settled law, and it does depend on degrees, but nevertheless it certainly can be eligible for copyright).
 
The prompt alone, no matter how creative it is, is not enough.
Direct quote
If a work’s traditional elements of authorship were produced by a machine, the work lacks
human authorship and the Office will not register it 26 For example, when an AI technology
receives solely a prompt 27 from a human and produces complex written, visual, or musical
works in response, the “traditional elements of authorship” are determined and executed
by the technology—not the human user.
Click to expand...

So I would worry if I used AI to produce entire code for me, but frankly I don't care, I am not using AI to write code and I couldn't care less, not my problem.
 
Last edited:
tomas said:
The prompt alone, no matter how creative it is, is not enough.
Direct quote


So I would worry if I used AI to produce entire code for me, but frankly I don't care, I am not using AI to write code and I couldn't care less, not my problem.
Click to expand...
Yes that's correct.

I doubt @BasS used a single prompt, and if there was a process of iteration and selection of what they liked then it is copyrightable.

I couldn't care less, not my problem.
Click to expand...
Well clearly you care a little bit, you made a whole incorrect post about it!
 
You are making assumptions on what he did and what he didn't. What he actually wrote is that he "asked Claude to make one". There is no mention of iteration of any kind, and even if he indeed iterated prompts, iteration alone is also not enough if the code produced by AI is used verbatim. You are entitled to your opinion and I have mine. And I think that your post is incorrect, not mine. Claiming copyright on AI generated code can bring on him other legal trouble. I added my 2 cents worth. Do what you want with it.
 
BasS said:
I too am a professional developer with 30 years experience. And tagging something with MIT “license” is essentially saying “there is no copyright”.

So I’m not exactly sure what “problem” you think there is.

it’s a good sounding reverb, pretty efficient with resources. Anyone can use it. That’s all there is to it
tomas said:
1. MIT license is NOT equal to "no copyright". "No copyright" is equal to public domain. You don't have any rights to AI-generated code, therefore you can't assign the license. Slapping an MIT License on it has no legal weight because there is no copyright to grant permissions for. The code is effectively in the public domain the moment it is created. You can't even put your name on it.

2. The code is not efficient on resources. As in typical AI-slop it is over-engineered. For example using 6th order Lagrange is over the top and waste of resources. AI did not ever hear any sound, it doesn't know what is audible and what is not.
Click to expand...

AI Generated code has whatever license YOU decide to use. Free, MIT, Commercial closed source, etc. Not sure where you got the assumption that this wasn't the case, however, you're incorrect. I'd suggest EVERYONE read the TOS, there are limits, like, some of the code you may write can be used by others, and your code could have been reused by code that was generated by other input. You can run into an issue(s) when you're using libraries that have other license. This is truly a muddy area, and can drive you crazy. Both Baas and tomas are correct here, like I said muddy waters. Myself I have my source MIT, you get the HEX file, however, I have to/should give you may Master Program if you ask for it.
Click to expand...


BasS said:
You're close on the MIT license,
Click to expand...
 
BasS said:
Hey everyone, I’m pretty new to doing Teensy projects, but getting really stuck in. I wasn’t super impressed with the stock reverb (designed for a different era with different hardware, not the 600 MHz DSP we have now!) so figuring this would be a well researched field with plenty of papers on how to design a nice reverb, I asked Claude to make me one. And I think it did quite a reasonable job. (I’m being modest here…)

I published it MIT licensed as an Arduino library at https://github.com/basscheffers/VERBtrix

If you want to hear a demo, here it is with the EP patch of an analog synth I am cloning: (dry first, then with reverb. Excuse my playing skills, or lack thereof.)


(Which will also be shared in due course)

Here it is on the prototype of the hardware for my synth:

View attachment 39850
Click to expand...
I've been working along a similar path in the past couple of weeks. I admit to using AI to greatly assist me in not only the code aspect but in terms of understanding how different algorithms and topologies work to create a particular type of reverb. I started out by asking for a Griesinger style algorithm and in a few hours of back and forth testing and tweaking by ear, I was asking to adapt what "we" had created into a clone of the famous PCM-70 "Rich Plate". I happen to own a PCM-70 and Claude suggested we use it to create some sweep files so it could compare the performance of a real PCM-70 to what we had created. I used Reaper and an audio interface to run the sweep files Claude had generated into the PCM-70. We used the real world data to tweak the algorithm. I think with a lot of human creativity and inspiration guiding it, Claude AI can be a powerful co-creating tool. After hours of fun making the various algo-style reverbs, I started exploring (again with Claude's assistance) a spring reverb emulation. I learned A LOT about how that works (a large string of short all-pass filters, like 100 of them in series). Pretty amazing stuff. And then I went on to model AN ENTIRE FENDER STYLE AMPLIFIER, including virtual triodes. In a couple of hours, all from my ideas and Claude's assistance. I know to a lot of you this is pure "cheating". But I can honestly say I have already realized sounds in DSP that Ive had in my head for decades and have tried many many times to get from analog hardware. In any case as long as we can resolve the potential legal debates, I'm really anxious to share my own work at some point. The more the merrier!
 
In my opinion, that's not 'cheating.' It's using technology where it makes sense. In fact, it would be unwise not to do so. If we didn't adopt or use new things, we wouldn't even have fire today, and we'd still be living in caves.
But I really fear that unless someone pulls the emergency brake on AI soon, things like this will be made exclusively by AI. Then our hobby will be dead. And not just that—a whole lot will change in our lives, and not necessarily for the better. But that's another topic.

Germany: "Under German and European law , copyright only protects 'personal intellectual creations' by human beings."
 
Last edited:
polaris26 said:
I've been working along a similar path in the past couple of weeks. I admit to using AI to greatly assist me in not only the code aspect but in terms of understanding how different algorithms and topologies work to create a particular type of reverb. I started out by asking for a Griesinger style algorithm and in a few hours of back and forth testing and tweaking by ear, I was asking to adapt what "we" had created into a clone of the famous PCM-70 "Rich Plate". I happen to own a PCM-70 and Claude suggested we use it to create some sweep files so it could compare the performance of a real PCM-70 to what we had created. I used Reaper and an audio interface to run the sweep files Claude had generated into the PCM-70. We used the real world data to tweak the algorithm. I think with a lot of human creativity and inspiration guiding it, Claude AI can be a powerful co-creating tool. After hours of fun making the various algo-style reverbs, I started exploring (again with Claude's assistance) a spring reverb emulation. I learned A LOT about how that works (a large string of short all-pass filters, like 100 of them in series). Pretty amazing stuff. And then I went on to model AN ENTIRE FENDER STYLE AMPLIFIER, including virtual triodes. In a couple of hours, all from my ideas and Claude's assistance. I know to a lot of you this is pure "cheating". But I can honestly say I have already realized sounds in DSP that Ive had in my head for decades and have tried many many times to get from analog hardware. In any case as long as we can resolve the potential legal debates, I'm really anxious to share my own work at some point. The more the merrier!
Click to expand...
Love this! I literally did the same thing with a Matrix 1000 synthesiser over the past few months. The Teensy version isn’t quite ready yet (mostly because I’m still working on hardware) but I published it as plugins for Mac.

Initial code was Claude analysing schematic and firmware and then tuning by my ear and prompts, and moving on to it having midi and audio access and the results are actually really good. It’s indistinguishable on some patches, not so much on others but they generally still sound good in their own way or I even prefer them!

The plugins are here: https://canunda.com/

Teensy project to follow later. :)
 
When my ESP32 port of the audio lib "is ready to use" we can try to port the plugins (with AI ;)) I'm excited to see how it goes.
 
Mcu32 said:
When my ESP32 port of the audio lib "is ready to use" we can try to port the plugins (with KI ;)) I'm excited to see how it goes.
Click to expand...
Oh cool! I was actually looking at doing a project with an ESP32-P4 and big screen interface; having audio library there will make it easier.
 
polaris26 said:
I learned A LOT
Click to expand...
As a learning tool AI is actually good. It really makes you progress faster in learning. And I think that learning is actually one of very few good uses of this technology.

What I am not comfortable with is calling entirely AI generated code as "my work" or "my ideas" or "my creation". AI generations are not (entirely) your ideas and your work. AI generations are the result of statistical prediction of next most probable word based on ocean of code written by **OTHER** humans, literally stolen from Github by AI companies, under false premise of "fair use". For what it is worth, "fair use"doctrine started as common law created by judges in the 18th and 19th centuries, and it was officially passed into statutory law by the U.S. Congress in the Copyright Act of 1976 when there was NO AI, no LLMs. "Fair use" was created for humans, for human consumption and human use. Applying "Fair use" to the act of training AIs to produce the same code is just wrong. That is why lawmakers say that AI generations have no copyright assigned to them, because AI code generations are literally statistical average of code that was fed to them and as such they belong to humanity, not one person, therefore they are public domain.

I know that I am in minority, but I get way more satisfaction in getting my own hands dirty and use my own brain to struggle with obstacles. It is called training. That is why people go to gym. Not to lose muscle. And I don't want to lose my brain just because AI will write the code faster and easier. There is some kind of pride that you can feel only when you actually make thru obstacles by yourself, not using helicopter.

Besides, I always chuckle when I see AI generated code full of subtle mistakes like for example using algorithm with 144dB SNR to feed 16 bit output instead of much cheaper that only gives 120dB and is twice as fast, and way more than enough. Or configuring I2S in left-aligned format instead of I2S format that codec actually uses by default. All these little things that show me that these all "greatest" and "frontier" models are not as good as advertised.

OK, end of my old man rant.
 
@BasS:
A lot has already been ported, but it's all still very messy, and here and there some details are still missing from what's already been ported.

I saw that the effect needs 256 KB of RAM, so we'll probably have to fall back on PSRAM (which won't exactly make it faster), and I don't know how fast the FPU is either. If necessary, we might have to switch to fixed-point integers there too. We'll see.
 
@tomas:
I've often seen that AI code tends to be better (even if you sometimes see some hair-raising stuff) than code written by many humans. The point is, AI rarely makes careless mistakes or misses small details.

Of course, those hair-raising bugs still need to be worked on and fixed together with the AI (and that's going to mean a good chunk of work for the ported lib).

It also has a positive side: it's fast.
 
tomas said:
As a learning tool AI is actually good.
Click to expand...
I’ve now spent two hours running some benchmarks, and I actually learned something quite surprising (well, for me). Float seems to scale really well (or rather, GCC does a really good job with Xtensa FPU).

An 8-channel mixer using float is not slower—or only marginally slower—than the same mixer using integers. With 8 channels+gains, 64-bit arithmetic is essential with fixed-point if you want good quality—or you might as well use floats.

Since the ESP performs so poorly with 16-bit anyway (no 16 bit dual extensions), I didn’t even bother trying that and instead use this fact to achieve higher quality at the same speed and implemented a n-channel float mixer.
 
Last edited:
Mcu32 said:
@BasS:
A lot has already been ported, but it's all still very messy, and here and there some details are still missing from what's already been ported.

I saw that the effect needs 256 KB of RAM, so we'll probably have to fall back on PSRAM (which won't exactly make it faster), and I don't know how fast the FPU is either. If necessary, we might have to switch to fixed-point integers there too. We'll see.
Click to expand...
Depending on what else is running, that may not be a problem. VA synths take almost no RAM, so a couple of effects like this (delays naturally take RAM too) plus synths might just work.

And don’t forget to ask Claude to optimise this, it will probably make some suggestions for trade-offs to make in return for lower RAM usage!
 
You must log in or register to reply here.
Back
Top