Project idea - DIY Google nest mini style speaker based on the HA voice schematics

It's an idea but I doubt (in its current form) that they've actually made this thing at JLC for example.

On the face of it, "four layer" board right? 

Only it's a double-sided four layer board which means a leap in the cost of assembly (it's not the sort of thing anyone should attempt at home since the parts on the first soldered layer need to be glued into place so they don't all drop off when the opposite side is soldered.

But then when I looked a little deeper it's a very high-precision board that's very densely packed and might have traces (I haven't checked) so narrow that its unlikely the low-cost board houses will even consider. Using tracks this narrow requires a high-precision board and that means moving to six or eight layers which starts to run expensive.

However I've got serious reservations that it would actually work in its current form. I've done over a 100 boards at this point so I'm getting better handle on what the low-cost guys can do.

I wonder if they've actually manufactured a prototype or just released the layout which looks very neat indeed? I noticed they've done firmware so perhaps they did?

In fact using four layers for this sort of thing is actually false economy as there are no ground "planes" which are pretty much a necessity at this sort of density with fast moving edges. It's not that it will fail EMI testing (this doesn't matter for non-commercial stuff) it's that with all those fields running riot there's an excellent chance that reliability will be compromised.

Looks like they're relying on copper pour for return paths and that (as the big names in this area like Eric Bogatin and Rick Hartley will tell you) is a no-no. We used to do this years ago and in fact, that's what I was taught back in the days of stone tablets. 😉 It's something I'm trying to unlearn as I tend do it without thinking.

When fast digital signals are involved the fields need something to couple to - hence it's better design practice to match a signal layer with a flat ground plane as close as possible. On a four layer stack (two and four layers are the hardest boards to get right ironically) layers two and three should be nothing more than solid copper with as few vias as possible. I saw a lot of other little "mistakes" (stuff we're all taught but is actually wrong) too.

Hence doing this sort of thing on four layers turns out to be worse than a simple false economy, but can actually result in something that simply won't work or won't work with any degree of reliability. Eric Bogatin's students redesign the Arduino Uno in his class and the results between the Open Source (the original - which works and sold in the millions) but done the right way the EMI improves almost by an order of magnitude.

When a manufacturer is making these things in the 1000s the costs drop off significantly but with all the will in the world, I don't see this as a practical home project, even for those will deep pockets as the total cost is likely to exceed that of buying a machine off the peg (as it were).

OK, so that's the bad news... but why not just take an existing device perhaps a used Amazon Alexa and upgrade the speaker and enclosure?

The speaker drivers are one of the most expensive single parts on the BOM (if not the most expensive) due to the complexity of making these things vs. an IC.

Everything up to the speaker is typically very high-quality audio (within the limits of the DAC of course) so it would make more sense to add a secondary amplifier and a speaker, would it not?

I'm just spit-balling of course, not trying to run this down.