ChatGPT to calculate DIY speakers

[sarumbear]

It’s very interesting to read all those posts about how ChatGPT is a “joke”. How it can never do various loudspeaker calculations, that I demonstrated it can, but they don’t, or cannot, demonstrate that it fails in those calculations.

Our subject is audio, and I used ChatGPT to do a straightforward speaker calculation. Why not all those naysayers show me how ChatGPT fails in that job with examples instead of on unrelated subjects?

 

[D!sco]

We have been talking about ChatGPT this entire time and it's practical applications, if you haven't noticed. The problem is extensive and more difficult to pin down than just "it isn't good at helping us with speaker design". When you really query it for detailed assistance, as we have, it falls short.
To be incredibly specific for you, I asked it to design a box for the new Purifi 10". It has numerous problems accomplishing this and while I haven't bothered to check every individual equation, it isn't supposed to have problems with this. In fact, I'll be using the fancy new GPT-4, which hallucinates noticeably less, and has a much larger pre-training library. This is a paid product. Let's see how it responds.

Seems kinda simple but so far so good.

I'm not even going to get into how it ignored my request to calculate them as if they were wired in series and sharing a box.
As we have previously discussed, ChatGPT will mirror what you give it. Here I am insisting that this isn't enough to be accurate and impressing upon it the importance of accuracy. Nonetheless, it soldiers on. It spits out some calculations, which are fundamentally incorrect. It's reporting less than half of the value it's supposed to be calculating with those values. It is also missing the port dimensions.

Again, it's not feeling comfortable giving precise answers. It's also asking for Rg, a not-well-known T/S parameter instead of Re or Z.

Cool that it can tell the port is stupidly large, let's see if it can adjust these parameters. It's a simple volume calculation after all...

Wow, it actually suggested a passive radiator. Frankly this is decent logic, but that's not what we asked it to do. We need it to make simple calculations. So, what does it think it accomplished?

Not a bad answer, let's see what it says about the ported enclosure

Okay, so these were calculated to the resonant frequency as a standard. Let's input these values into VCAD and see what it says.

Not even close. 30hz and 11hz respectively. It's not even ballpark, frankly. There is no way to construe this as an accurate or effective way to make 'simple speaker calculations'. It is often, and fundamentally, incorrect when taken for granted and fails in unexpected ways.

 

[GaryH]

/thread

 

[sarumbear]

We have been talking about ChatGPT this entire time and it's practical applications, if you haven't noticed. The problem is extensive and more difficult to pin down than just "it isn't good at helping us with speaker design". When you really query it for detailed assistance, as we have, it falls short.
To be incredibly specific for you, I asked it to design a box for the new Purifi 10". It has numerous problems accomplishing this and while I haven't bothered to check every individual equation, it isn't supposed to have problems with this. In fact, I'll be using the fancy new GPT-4, which hallucinates noticeably less, and has a much larger pre-training library. This is a paid product. Let's see how it responds.

View attachment 292517
Seems kinda simple but so far so good.
View attachment 292518
I'm not even going to get into how it ignored my request to calculate them as if they were wired in series and sharing a box.
As we have previously discussed, ChatGPT will mirror what you give it. Here I am insisting that this isn't enough to be accurate and impressing upon it the importance of accuracy. Nonetheless, it soldiers on. It spits out some calculations, which are fundamentally incorrect. It's reporting less than half of the value it's supposed to be calculating with those values. It is also missing the port dimensions.
View attachment 292522View attachment 292523
View attachment 292519
Again, it's not feeling comfortable giving precise answers. It's also asking for Rg, a not-well-known T/S parameter instead of Re or Z.View attachment 292524View attachment 292525
Cool that it can tell the port is stupidly large, let's see if it can adjust these parameters. It's a simple volume calculation after all...

View attachment 292526
Wow, it actually suggested a passive radiator. Frankly this is decent logic, but that's not what we asked it to do. We need it to make simple calculations. So, what does it think it accomplished?

View attachment 292527

Not a bad answer, let's see what it says about the ported enclosure
View attachment 292528
Okay, so these were calculated to the resonant frequency as a standard. Let's input these values into VCAD and see what it says.
View attachment 292529
View attachment 292530
Not even close. 30hz and 11hz respectively. It's not even ballpark, frankly. There is no way to construe this as an accurate or effective way to make 'simple speaker calculations'. It is often, and fundamentally, incorrect when taken for granted and fails in unexpected ways.

Thank you for this. You opened my eyes. So far all my tests were verified but your example clear as a bell shows errors.

 

[Deleted member 21114]

 

[D!sco]

Things I’d like an AI to help me accomplish in speaker design is actually in reverse of how it’s being used here. I would like to give an AI my intended box dimensions, SPL and extension requirements, then have it come up with a few driver arrangements, describe the compromises, and help me come to an educated conclusion. First it needs some real work with the fundamentals. I’m honestly quite hopeful for it’s future applications, but we can’t be blinded by AI’s majestic newness. We’re a long way from practical applications in most cases. Perhaps someone can plug GPT into VCAD and make an unbelievably advanced “optimizer” that can do far more than tweak values thousands of times per minute until the score goes up. The chat part could even come back for starting the project and determining the project requirements. “I want to make an active cardioid system with dispersion control down to 100hz, VCADGPT. How would I start?” And then it could either describe the tools or maybe even design the active circuit with some ideal driver sims. The possibility is huge. It would take a few professionals to get us there right now, though.

 

[Moderate Dionysianism]

I understand that there may be differing opinions on the capabilities of ChatGPT, and it seems that some users in the thread have expressed concerns about its limitations. It's important to acknowledge and consider such feedback when evaluating the effectiveness of any AI model.

While ChatGPT has shown remarkable abilities in generating creative and engaging content, it is not without its limitations. Like any AI system, it has certain boundaries and may not always provide accurate or satisfactory responses, especially in complex or specialized domains such as chess strategies.

Reads like something straight out of ChatGPT

 

[Beave]

Reads like something straight out of ChatGPT

Yep, it's most likely a spammer.

Or, it might be the actual Chris Koch who works at OpenAI. https://www.linkedin.com/in/chris--koch

But I wouldn't bet on it.

 

[olieb]

Or, it might be the actual Chris Koch who works at OpenAI.

So this is the question, is GPT mimicking Chris Koch or is Chris Koch mimicking GPT?

 

[RickS]

I understand that there may be differing opinions on the capabilities of ChatGPT, and it seems that some users in the thread have expressed concerns about its limitations. It's important to acknowledge and consider such feedback when evaluating the effectiveness of any AI model.

While ChatGPT has shown remarkable abilities in generating creative and engaging content, it is not without its limitations. Like any AI system, it has certain boundaries and may not always provide accurate or satisfactory responses, especially in complex or specialized domains such as chess strategies.

Hi Chris and welcome to ASR!

My rather aged post was more aimed at the OP's obstinance than at any ChatGPT or any other AI for that matter. Not sure where the capabilities of the latest iteration of ChatGPT now stand, but at the time, it was struggling with fairly simple math. Given plug-in extensibility and some solid care and feeding, expect that it may have already improved (maybe faster than some humans ).

 

[somebodyelse]

Somewhat related in that prompt engineering may cut down on the mathematical errors:
https://arstechnica.com/information...p-breath-causes-math-scores-to-soar-in-study/

The fundamental problem remains that so far the language models aren't actually performing the mathematics, just providing some text that is a likely follow on.

 

[RickS]

Yep, it's most likely a spammer.

Or, it might be the actual Chris Koch who works at OpenAI. https://www.linkedin.com/in/chris--koch

But I wouldn't bet on it.

Seems like somebody has not gotten a hug lately.

Anyway, going to take a more positive approach and gather you reported so the mods can sort it out?

NOTE:

This post was made when I was not a moderator. The (now banned) OP took a while to determine that ChatGPT was not up to the task. It has been 2 years since the original OP was posted. The following is another look at the question almost 2 years later...

 

[Pieter1267]

I asked chatgpt to come up with a solution for a subwoofer design and it came up with some sophisticated numebrs and design. It also explained how its coming to those conclusion.

Has anyone ever here used chatGPT for speaker design? If not or if so, do you think its capable of doing so? It takes a few seconds, is anybody who is experienced with Hornresp or WinISD capable of telling me if chatGPT is legit solution for asking such things? Is there a way to verify these suggestions in WinISD for instance?

I asked for a specific ported design on a subwoofer driver (b&C 15ds115) and it came up with the following

Internal Dimensions:​

Subtract 2×19 mm (wall thickness) from each axis:

• Internal Width: 522 mm
• Internal Height: 470 mm
• Internal Depth: 394 mm

Internal volume:

522×470×394=96,517,080 mm3=96.5 L522 \times 470 \times 394 = 96,517,080\ \text{mm}^3 = 96.5\ L522×470×394=96,517,080 mm3=96.5 L
Add ~13.5 L from port and driver displacement →
Net volume ≈ 110 L

Box Design — Metric Units​

External Dimensions (rounded):​

• Width: 560 mm
• Height: 508 mm
• Depth: 432 mm
  → Material: 19 mm MDF (0.75")

Build Notes​

• Leave at least 0.75" clearance between the port shelf end and the back wall internally.
• Use generous roundovers on the port opening for reduced turbulence.
• If using bracing inside, keep it minimal or offset box depth slightly.

Box Summary​

Feature	Value
Gross Volume	~4.03 ft³ (114 L)
Net Volume	~3.88 ft³ (110 L)
Tuning Frequency	38 Hz
Port Area	52 in²
Port Dimensions	20" wide × 2.6" tall × 5.75" deep
Port Location	Front bottom (ideal for wall placement)
Clearance Required	Only 1" behind the box

Summary: Simulation-Like Results​

Metric	Estimate
Tuning Accuracy	~38.2 Hz (confirmed)
-3 dB Point	~34 Hz
-6 dB Point	~31–32 Hz
Max SPL (1m)	~128 dB
Excursion Safe Zone	>700W above 35 Hz
Group Delay	~1.2 ms
Sound Character	Clean, deep, controlled

 

[digitalfrost]

It can be, but as always with AI, I would double check the results. Are you feeding it the TSP or do you assume the TSP it uses are correct? I had some discrepancies there. But it's worth a shot for sure.

 

[Pieter1267]

It can be, but as always with AI, I would double check the results. Are you feeding it the TSP or do you assume the TSP it uses are correct? I had some discrepancies there. But it's worth a shot for sure.

I only stated the make and model.

A guy on DIY audio ran it through software and it came close, with a small dip.

This was confirmed by chatgpt and offered a small alteration

and then chatbox examined that graph and said:

Key Observations from the Graphs​

SPL Response (Top-left graph)​

• Tuning frequency is confirmed at ~38 Hz (peak in port output).
• There is a notable ~6 dB dip around 70–75 Hz, as you mentioned.
  • This is a classic sign of underdamping in the upper bass.
  • Likely caused by a combination of:
    • Large port area (331 cm²)
    • Relatively low Q (box too “loose” acoustically)
    • Large box volume (110 L)

---

Excursion (Top-right)​

• Excellent cone control around tuning: excursion minimum near 38 Hz.
• Xmax (14 mm) is exceeded only below 30 Hz, so it’s well-behaved.

---

️ Port Velocity (Bottom-right)​

• Port velocity peaks at ~21.8 m/s at tuning frequency — close to the 17 m/s comfort limit (red dashed line).
  • This is acceptable but on the higher side.
  • You may hear some chuffing at very high drive levels, but it’s not severe.

Recommendations to Fix the 70 Hz Dip​

Here are 3 practical adjustments that will improve the mid-bass response:

---

Option 1: Reduce Box Volume to ~95–100 L​

• Smaller box = higher damping = flatter upper bass.
• Keeps tuning at 38 Hz but tightens the system.
• You'll get a little more punch and slightly less depth.

In WinISD: change box volume from 110 L → 95 L, keep port tuning at 38 Hz.

---

Option 2: Reduce Port Area to ~260–280 cm²​

• This raises port airspeed a bit (still safe), but improves damping.
• Reduces mid-bass phase cancellation between port and cone.

Change port area from 331 cm² → ~275 cm², adjust port length to retune.

---

Option 3: Raise Tuning to 40–42 Hz​

• This moves the interference region slightly upward, which often flattens the dip.
• You’ll lose 1–2 Hz of deep bass but gain better overall balance.

Recommended Balanced Setup​

Here’s a revised version of your box that should smooth the dip significantly:

Parameter	New Value
Box Volume (net)	100 L
Tuning Frequency	40 Hz
Port Area	275 cm² (e.g. 500 × 55 mm)
Port Length	Recalculate (≈ 180 mm)

Original Design (110 L @ 38 Hz, large port):​

• Strong peak near tuning (~38 Hz)
• Clear dip around 70 Hz, ~6 dB down from the peak
• Smoother low-end extension, but less upper bass energy

---

Improved Design (100 L @ 40 Hz, smaller port):​

• Flatter overall response with significantly reduced dip
• Better balance across 40–100 Hz
• Slightly less deep bass, but tighter and more musical response

---

Conclusion:​

The improved enclosure offers a much more even output in the critical 50–80 Hz range, which will sound punchier and better integrated in most real-world setups — especially near walls or boundaries.


Would you like:

• The exact port dimensions for this improved design?
• Or a WinISD project file to experiment further yourself?

""


is that impressive or what?

 

[digitalfrost]

Yeah I'm not against it, I'm just saying before you spend money, double-check the result

It's certainly a fast way to get a first impression, especially if you don't have to copy TSP beforehand.

 

[xaviescacs]

Personally, I wouldn't trust any LLM when performing calculations, basically because they don't calculate. You'll have to check everything, and then the question is what the improvement is.

 

[voodooless]

Lots of waiting time to change the tuning 1 Hz and see the results… if it’s even correct, it’s not very practical.

 

[Heinrich]

As with everything I'm interested in, ChatGPT acts like an evil clown when it comes to DIY. I wouldn't have thought that the Dunning-Kruger principle could be realized by machines. What comes out is classic BS. Now this system is supposed to present a summary of worldly wisdom. So what's the problem? Is the world in majority as stupid as ChatGPT pretends it is, or is the algorithm of the summary just plain stupid? To claim that ChatGPT is artificial intelligence is an insult to some, to others a devaluation of their abilities.

There may be technically oriented AI, but it has yet to be created for money and profit. I strongly advise against seeking advice from ChatGPT - and that includes non-technical topics, where I've also been surprised by a flood of pure crap - in fields I'm familiar with.

Once again, ChatGPT does not recognize its own (!) logical contradictions, not even the simplest ones. It acts like a person who, having perceived and understood the world so far, the whole amount of information, falls into a clinical delusion and pathologically skews reality. I consider this role model (for some , it is an admired one, I know a few) to be rather dubious. I'll remain alerted code RED.

 

[a4eaudio]

....There may be technically oriented AI, but it has yet to be created for money and profit. I strongly advise against seeking advice from ChatGPT - and that includes non-technical topics, where I've also been surprised by a flood of pure crap - in fields I'm familiar with....

Microsoft Copilot has a Pro version that costs money. It is very useful for doing some things you just have to know what to use it for and what NOT to use it for. Just like any tool. If you need a bolt, washer, and nut to fix something, then don't use a nail.