ChatGPT to calculate DIY speakers

[kemmler3D]

I don’t understand why you are commenting? Do you have any other purpose other than pollute the thread and confuse the subject?

The first sentence of this thread says:



Bad forum etiquette!

I was responding to your statement about GPT being a useful assistant. My response: It's not! And I further explained why I think so. Understand?

 

[kemmler3D]

An assistant in engineering and IT. It’s stated at the very beginning, the title has the word calculate on it. You have ignored both of them.

All I understood is that you are an ignorant person who posts for the sake of posting.

As others have explained, it's just as bad for engineering and IT as it is for anything else, for the same reasons that it's bad for my work. It's inherent to the limitations of the technology as it exists today.

 

[sarumbear]

As others have explained, it's just as bad for engineering and IT as it is for anything else…

Thankfully others can understand the usefulness of ChatGPT as an assistant.

 

[kemmler3D]

Thankfully others can understand the usefulness of ChatGPT as an assistant.

What is describe there is a good use case because validation of the output is built into the process, and it can actually save you time. No argument there.

My larger point is that it's not good for any situation in which you need an answer that isn't somewhat self-validating (like code) or you don't already know the answer.

For example I just asked it to design a crossover circuit... it contradicts itself in the same response.

1. Choose filter types: There are several types of filters that can be used in a crossover circuit, including first-order (6 dB/octave), second-order (12 dB/octave), third-order (18 dB/octave), and fourth-order (24 dB/octave). For this design, we will use a second-order crossover with a 12 dB/octave slope.
2. Determine crossover components values: For a second-order crossover, you will need a capacitor and an inductor for each driver. The capacitor is used to block low-frequency signals from reaching the tweeter, while the inductor is used to block high-frequency signals from reaching the woofer. The values of the components are calculated using the following formulas:
   Capacitor value (in microfarads) = 1 / (2 * pi * crossover frequency * tweeter impedance) Inductor value (in millihenries) = 0.159 / (crossover frequency * woofer impedance)
   Assuming a tweeter impedance of 8 ohms and a woofer impedance of 4 ohms, the capacitor and inductor values for a crossover frequency of 2,500 Hz are:
   Capacitor value = 3.3 microfarads Inductor value = 1.3 millihenries
3. Assemble the crossover circuit: Once you have determined the values of the components, you can assemble the crossover circuit. Place the capacitor in series with the tweeter and the inductor in series with the woofer. Connect the two branches of the crossover in parallel to the amplifier output.

So which is it? One component per driver, or two? I'm no EE and maybe this is technically right... but it seems wrong, even self-contradictory, and it's definitely confusing. If it's wrong, and I don't know that, I'm going to put together a really s***y speaker, best case scenario. Worst case scenario I cost myself $100s by destroying drivers because I used ChatGPT and not a normal crossover calculator.

 

[Axo1989]

In short, ChatGPT knows more about speaker design then most people on this forum, or on other forums and more than many at some speaker manufacturers, hence it is a tool that works. That is what I said and I stand by it.

I am not a LLM specialist. I don't care how the results are reached. I'm an engineer and I design websites using PHP. For me the results I am getting are correct and saves me time and effort. Every tool have its deficiencies. You can get a wrong answer on Excel too.

Haha, I feel the same way about the tiny elves inside my loudspeakers. Such fine musicians !!

 

[Axo1989]

Thanks, this helps...

Happen to understand how Amazon's Alexa AI technology compares? I often use Alexa for simple metric to standard conversions, but found a case last year, where it was close but not exactly right.

No, I haven't looked at specifics for Alexa or its APIs. They are quite different under the skin from the GPT large language model though.

 

[xaviescacs]

No, I haven't looked at specifics for Alexa or its APIs. They are quite different under the skin from the GPT large language model though.

One could argue that speech recognition systems are more "intelligent" because they need to "understand" what we say and take concrete actions that affect the physical environment, just like an animal does. Also, speech recognition systems are much more complex, as they need to perform different tasks: process the input, decide if the message received is clear and unambiguous enough, execute the corresponding task, give feedback, etc. To do so they need several AI techniques to work in combination. Chat GPT and other LLM essentially only do one thing, it's more about brute force.

So yes, the guts are very different, although they both use deep neural networks at some point, but who doesn't xD

https://towardsdatascience.com/how-...o-natural-language-processing-ai-7506004709d3

 

[Rick Sykora]

While I can see potential of an AI for speaker design, for me it would have to exceed (graphing, driver database, etc.) what I can do with a free app like VituixCAD. Like so...

Otherwise, the OP is an interesting test, but not much else.

 

[sarumbear]

While I can see potential of an AI for speaker design, for me it would have to exceed (graphing, driver database, etc.) what I can do with a free app like VituixCAD. Like so...

View attachment 282026

Otherwise, the OP is an interesting test, but not much else.

May I see how one can answer the 2nd question using those programs, other than multiple iterations?

Question to ChatGPT:
What will be the approximate range of TS parameters for a driver to have f3=20Hz on a 200L vented box and what will be the port length for a 3" pipe?

 

[Deleted member 21114]

This might illuminate.

 

[xaviescacs]

While I can see potential of an AI for speaker design, for me it would have to exceed (graphing, driver database, etc.) what I can do with a free app like VituixCAD. Like so...

View attachment 282026

Otherwise, the OP is an interesting test, but not much else.

AI is about algorithms, topologies and optimization problems. I guess all three things are already used in speaker and driver design.

 

[FrankW]

I was responding to your statement about GPT being a useful assistant. My response: It's not! And I further explained why I think so. Understand?

ChitGPT?

 

[Rick Sykora]

May I see how one can answer the 2nd question using those programs, other than multiple iterations?

Question 2 is not a common speaker design problem, so takes a different approach to help solve. I would use BassBox to do, but there are other free tools that can help…

The main use case is for replacing a woofer in an existing cabinet. If you do not know the original driver parameters, the process is an inexact one and likely will require iteration as it is unlikely you will find a woofer that has the exact parameters anyway. For example, Parts Express offers a free tool to search for potential driver candidates.

In any case, I already acknowledged the potential for AI to assist in speaker design. ChatGPT is not close to being capable or trustworthy enough for my purposes, but if you are happy, have at it.

 

[sarumbear]

Question 2 is not a common speaker design problem, so takes a different approach to help solve. I would use BassBox to do, but there are other free tools that can help…

If you are a speaker designer working for a company you are most often tasked to design a speaker where marketing defined the physical parameters. In which case the question is the actually most common one. Don’t you think?

BassBox is good for the hob, however it requires you to select from the alignments offered, not directly gives you the parameters.

 

[tomtoo]

If you are a speaker designer working for a company you are most often tasked to design a speaker where marketing defined the physical parameters. In which case the question is the actually most common one. Don’t you think?

BassBox is good for the hob, however it requires you to select from the alignments offered, not directly gives you the parameters.

Hopefully professional bridge designers dont trust chatgpt blinde. With bad speaker deasigns i have no problem. Iam used to them.

 

[kemmler3D]

If you are a speaker designer working for a company you are most often tasked to design a speaker where marketing defined the physical parameters. In which case the question is the actually most common one. Don’t you think?

I have been that marketing person, and... this is a good point.

As it stands today, I still personally wouldn't trust the results of a GPT calculation unless I verified it with a traditional calculation.

If, someday, we can be sure that ChatGPT (or other LLMs) are incapable of getting math wrong (like our old trusty spreadsheets) then that would be quite powerful. The use case here (feeding it arbitrary optimization problems in plain english) is really major, no argument about that.

 

[davide78]

If, someday, we can be sure that ChatGPT (or other LLMs) are incapable of getting math wrong (like our old trusty spreadsheets) then that would be quite powerful. The use case here (feeding it arbitrary optimization problems in plain english) is really major, no argument about that.

There are mathematical problems that computers cannot solve, no matter how poweful and complicated they are. This has been known since the 1920s, from the works of Turing, Goedel and others. Solving an arbitrary optimization problem in plain english is one example of such "undecidable" problems. So, language models will always get some math wrong and some code wrong. They can produce good results for problems that we already know how to solve, and for which there is enough information in the data set. Ask them to solve a new problem, and you most likely get a wrong answer.

 

[kemmler3D]

They can produce good results for problems that we already know how to solve, and for which there is enough information in the data set.

Right, I'm not saying there's some kind of superior mathematical ability in these things, I just mean "when they stop making basic errors in calculation" - you know, normal arithmetic, algebra, maybe a little calculus...

 

[Rick Sykora]

If you are a speaker designer working for a company you are most often tasked to design a speaker where marketing defined the physical parameters. In which case the question is the actually most common one. Don’t you think?

BassBox is good for the hob, however it requires you to select from the alignments offered, not directly gives you the parameters.

I want some woofers that fit the box (volume) you specified. Parameters alone would not even ensure the driver was feasible. That is what the Bassbox wizard would help you find...

Lol, ChatGPT changed your f3 to a box tuning frequency. That already would cause the results to be incorrect less optimal. Note I changed the f3 range in BassBox as at 20 Hz, no drivers matched.

Did you try to get ChatGPT to find woofers that match the results it gave?

 

[sarumbear]

Lol, ChatGPT changed your f3 to a box tuning frequency. That already would cause the results to be incorrect.

I’m sure you know what Butterworth alignment is but for some reason you pretent that you don’t know.