Could AI design the ‘perfect’ speaker (or any other audio component !)

[xyvyx]

You would first need to define what is a perfect speaker... that's not possible imo

Nah, that's easy. A device/cone that's able to move in perfect response to an electrical input signal.
It's constraints are that it only needs to be able to do this between ~ 20 - 20khz or thereabout.

 

[kemmler3D]

A device/cone that's able to move in perfect response to an electrical input signal.
It's constraints are that it only needs to be able to do this between ~ 20 - 20khz or thereabout.

I think everyone can agree on perfect frequency + phase response, but:

What SPL does it need to produce?
What about subsonics?
What dispersion pattern does it need to produce? Should it have a cardioid, dipole, omni, or other pattern? Same or different pattern for vertical / horizontal?
How much distortion is acceptable at any given frequency / SPL ?
What should it cost?
How much should it weigh?
How big should it be?
What are the efficiency / power requirements?

These are all things where you will find differing opinions and none are definitively correct.

 

[Digital_Thor]

Also preferences. Some people like the sound of wider dispersion speakers, because they 'illuminate' the room. Whereas other prefer a more focused sound image in front of them, with narrow dispersion speakers and more damped rooms.
So, perfect speakers for announcements on a train station or perfect speakers for your head phones.... Or the one saying 'beep' on your motherboard, during a successful boot.

Where almost back to a discussion of finding the best driver for a given speaker project. People often fall in love with a driver, then build the speaker. Rather than figuring out which sound they like, and then choose the drivers fit for the purpose.

And all those speakers presented by AI to this point, mainly seem to arouse a visual feeling, rather than a performance goal - especially one we don't already know off.

Visual eye candy is nice - in many areas.... But performance usually last longer.

 

[RDoc]

As has been mentioned, AI, at least systems like ChatGPT and MidJourney, are really just putting together words or pixels it's been trained go together. To design a "perfect" speaker I think you'd need something more like a special purpose mechanical simulator, more like a fluid flow simulator with a multi-dimensional optimizer.

The idea would be to tell the system what you're current definition of "perfect" was in terms of things like price, size, various acoustic behaviors, etc. in terms it was set up to understand along with telling it what's most important in design trade offs. Then, probably with a few starting designs, it would iterate through a bunch of designs, varying elements of the design, then emulating each for how well it met the input parameters. By optimizing across all the design factors it would get to a best design. It would have to try everything, all the speaker elements, the enclosure, dsp corrections, etc.

Not trivial to do at all but probably not impossible.

 

[levimax]

I am not an AI expert but since no human can beat a computer at chess since 1997 I don't think it is a stretch to conclude that as long as the definition of a "perfect speaker" and good data on available components were provided that the correct type of AI could, through brute force and persistence if nothing else, come up with a speaker design that is better than anything a human designer could come up with. At that point the speaker would have to be built and tested and the results fed back to the computer for a couple rounds of optimisation just like the normal design process but in the end I think the computer design would win. Since I am speculating about all of this I will continue that with current technology it is probably too expensive to do it this way but that will probably change soon.

 

[RDoc]

IMHO using AI or any computer system to design a speaker is completely dependent on how good the speaker simulator is.

 

[Barrelhouse Solly]

Humans create the software. Humans provide the data. Humans set the specs. Creative work would be an interesting development.

 

[Yorkshire Mouth]

Some of the comments here are hugely ignorant as to what AI might be used for in speaker design, at least initially.

There's a whole dollop of bias, too.

In the first instance, AI wouldn't be told to "create the perfect speaker", and off you go.

It'd be told to create a speaker with less than x% distortion, FR within +/-x%, and told of the factors relating to reflections, etc. It'd also be told what techniques have been tried previously. It'd then be given a price profile.

It can work on numerous levels, but in this respect, in this context, it's just cutting out a whole lot of time and waste.

Whether it's speaker design or whatever, AI can just use past experience to figure out what's worked and what hasn't, and use that information to inform the likely outcomes of future designs using those techniques.

 

[Emlin]

Some of the comments here are hugely ignorant as to what AI might be used for in speaker design, at least initially.

There's a whole dollop of bias, too.

In the first instance, AI wouldn't be told to "create the perfect speaker", and off you go.

It'd be told to create a speaker with less than x% distortion, FR within +/-x%, and told of the factors relating to reflections, etc. It'd also be told what techniques have been tried previously. It'd then be given a price profile.

It can work on numerous levels, but in this respect, in this context, it's just cutting out a whole lot of time and waste.

Whether it's speaker design or whatever, AI can just use past experience to figure out what's worked and what hasn't, and use that information to inform the likely outcomes of future designs using those techniques.

I'm sorry, Dave. I'm afraid I can't do that.Dave: What's the problem?

 

[D2013]

Some of the comments here are hugely ignorant as to what AI might be used for in speaker design, at least initially.

There's a whole dollop of bias, too.

In the first instance, AI wouldn't be told to "create the perfect speaker", and off you go.

It'd be told to create a speaker with less than x% distortion, FR within +/-x%, and told of the factors relating to reflections, etc. It'd also be told what techniques have been tried previously. It'd then be given a price profile.

It can work on numerous levels, but in this respect, in this context, it's just cutting out a whole lot of time and waste.

Whether it's speaker design or whatever, AI can just use past experience to figure out what's worked and what hasn't, and use that information to inform the likely outcomes of future designs using those techniques.

Part of the issue is that a lot of stuff that is attributed to "AI" isn't even "Ai". It's stuff like max, min, & root finding algorithms that are iterating through multiple dimensions.

Ai is nothing more than big techs next big buzz word. It's used and abused just like its predecessor was "the cloud"!

We had a long discussion about this last week at work (all developers)!