Working in tech, everyone has been getting excited about using it to make our jobs easier.
When I first tried it, I was really excited; I used it to write a summary of a software feature that I use frequently.
Then I asked it to write a summary of something else, and what it spit out was nearly identical to the first request.
For someone that gets paid to demonstrate some semblance of expertise, ChatGPT is a really dangerous tool for me to use. If I don't understand the subject thoroughly, I can wind up sending out a doc or an email that has information that's just completely wrong.
Am not quite sure how Chat GPT really is working. However generative models are very close to how human brains work. Children learn by hearing and copying. They then create a knowledge map for further interaction with the world. Basically they create known safe generation of output from input that they use when grown up, that's until you mess with a humans emotions does it all become undone and they generate from all sets. Chat GPT is working just like a new learner or a human who is not yet cultivated to anchor the thought process or check their output generation.
Chat GPT has been quite accurate on making suggestions for this two way build. These woofers have breakup at 3kHz
Children learn to lie better and better and better, until they invent the white van speaker scam but with Dayton drivers.
I will post pictures soon, once I figure out what non resonant bracing to use. With what I have right now, the speaker walls are well damped when you do a knuckle test it made me feel warm inside. However I happened to knuckle test the bracing itself and its resonant 🙂 probably because it looks like a tuning fork😆, but the walls are dead silent, will keep you updated
However, when chatGPT or similar take over what we learn, we stop learning.
In other words, the risk is that this is a next step in robotizing human beings.
I have always had this idea of modeling humanity and simulating the future to see how we unravel 🤣
For the box dimensions am using prime numbers 11" * 13" * 17" with a U spring brace on the larger sides like shock absorbers at 7" am liking it so much I may not use polyfill but as usual its expected to provide some high frequency damping although the braces introduce diffraction inside the cabinet. The woofer is center on the 11" * 17" baffle with tweeter and port on opposite sides with center's 6" from woofer center. The port is 2" diameter and can move air smoothly. Also this is not a high power design subwoofer box with a driver trying to tear all things apart. 😉 Feel free to comment, am currently comfortable with that single spring brace and kinda prefer the bass performance without stiffer bracing for this particular situation but the project is far from finished
ChatGPT may have hallucinations but AI is advancing at a rapid pace, I wonder how far away we are from an AI generated crossover design. I don't think it's unrealistic that a company could leverage AI within a year or two and release software powered by AI.
I would love to put my speakers in a box. Make driver measurements and send it to an AI to do the crossover work. Easy peasy
I would love to put my speakers in a box. Make driver measurements and send it to an AI to do the crossover work. Easy peasy
VituixCad can already do that; just tell it what response shape you want for each driver, and it will do it.
I think what would be really neat would be some software program where you could tell it the crossover point, and then let it spit out a hundred different crossovers with varying slopes and Z locations.
This wouldn't require AI though; it's just a brute force way to automate the crossover process.
I think what would be really neat would be some software program where you could tell it the crossover point, and then let it spit out a hundred different crossovers with varying slopes and Z locations.
This wouldn't require AI though; it's just a brute force way to automate the crossover process.
You still need to manually select a handful of promising candidates and listen to them (with a real-time emulator).
Everything starts with a well-restricted and properly working set of acoustic target functions. The definition of these sets requires experience and knowledge of the detailed acoustic situation and the drivers and which aspect we want to have priority etc. All in all a lot of experience and knowledge. That's where AI could offer some help to weed out the crap.
Once you have those of target functions the rest can be fully automated to obtain the correction functions for the drivers. Then for the final synthesis of the analog/passive circuit again a lot of knowledge and experience is needed to make an educated guess for a starting point and AI might help there, too.
As a semi-retired engineer "AI" has popped here and there in various forms all my working life. Neural networks have been used for decades to optimize the design of networks like a speaker crossover. The approach tends to be inefficient compared to others and has not been widely used but it has been around and promoted by AI enthusiasts for a long time.
The speaker build was very interesting and is still a work in progress. I had one type of brace that made the walls very inert. Tapping the walls did not produce any external ringing. Inserting the driver and tapping on it produced no ringing and was very pleasant. Then I removed the driver and tapped on the brace and it was ringing, at that time I didn't fully comprehend the potential of that brace and changed it, the new brace didn't ring, the walls weren't as inert as before and needed an extra brace. I replaced the speaker and tapped again and the sound was more boxy. Later I realized the magic of that ringing brace and the internal cancellation that was taking place with the driver inserted.