Yeah, the “this looks somewhat reasonable (later) eh wait does it?” happens all the time on this newfangled interweb thing 
I spent a bit more time googling but failed to find any single-transistor design that convinced me that the author had any idea what they were doing, or for that matter had done any testing whatsoever… But I’m sure it’s somewhere out there, there’s just too much noise in the way.
But until I find one, I think I’d recommend, in order:
- getting some cheap computer speakers, which tend to have good enough amplifiers and fairly robust input stages
- getting a cheap “class D” amplifier board, e.g. https://www.adafruit.com/product/2130
- getting a slightly cheaper (but much worse) prebuilt LM386 board (typically from ebay/aliexpress)
or if you’d really like to DYI from scratch,
-
build yourself a Noisy Cricket (or some of its predecessors, the ancestry of which you can trace all the way back to the LM386 datasheet)
(more here: http://beavisaudio.com/projects/noisycricket/, including a slightly enhanced version of the schematics)
(Note: The MPF102 is no longer in production, so can be tricky to source. Other small-signal JFETs probably work just fine. Or you can use an opamp. Or you can drop the input stage completely, if you don’t plan to use it with a high-impedance source like e.g. a guitar mic.).
I’m also not sure why connecting an 8 ohm speaker directly to a GPIO pin doesn’t damage the pin; sure, there’s some internal resistance that’ll cause the voltage to drop, but seems you’d still exceed the 40 mA max output. Oh well, I’ll figure it out some day.