The audio was mono, sounded terrible, and only included a 60 second sample of a song.
So in a few decades we've added stereo, video, and ... how long are TikTok samples, again?
(The whole project reminds me of a colleague who brought his childhood programs into modern archival format by using a walkman to play the cassettes into the audio-in of a Sun, then decoding from there into plain ascii.)
After more thought it's unclear to me whether this is actually storing PWM. It can make sense to emit PWM even if what you've actually stored is something quite different, because turning PWM into recognisable sounds with cheap electronics is easy.
The PWM described (~8 MHz of one bit PWM) ought to be capable of far better fidelity than we see in Techmoan's video or the linked audio from the blog post - it's pretty similar to (mono) Super Audio CD for example.
So that's why I end up thinking that the interesting part is probably in the chip you can't easily get at. Like maybe it's MPEG audio layer II with a PWM decoder or something.
It's probably a simplistic codec that was originally designed for speech and not music, which would explain the crappy audio.
There are a bunch of specialised microcontrollers designed for toys with like a 4bit CPU, a few GPIOs, 64 bytes of ram and a big massive mask ROM (like a few hundred KB) to hold audio data. They are designed for toys were you press a button and the doll speaks a random catchphrase. One of the common chips in the era was the Winbond "PowerSpeech", which used 4-bit ADPCM at 6khz, 8khz, or 12khz
I think Tiger just took one of these MCUs, picked the biggest mask ROM available and put a single 60 second music clip on it.
It's always fascinating to read about people reverse engineering obscure tech like this. I don't even know where to start to acquire the skills to do something like this.
Start by processing formats for which you have docs. Then try it referring to the docs as little as possible. Then move to redpointing formats.
(Feynman said something to the effect that if one is always trying to solve puzzles, and always checking to see how other people solved them, then one day one will discover that one has solved something no one else already had.)
Honestly the first and sometimes last step is to just not be afraid to break things.
If the author had been as afraid as me of reverse biasing the circuit, he might not have gotten started. Try things, and if something breaks there's still something you can learn from that behavior.
I was just about to comment with the suggestion of adding a low pass filter for nyquist transformation, but reddit beat me to it!
Still amazing that it's a self-contained audio device, no actual addressing or protocol needed. It would be interesting if (as the author thinks) these were custom ASICs with built in songs from the factory, or some combination of EEPROM and decoder/driver chip. Does anyone remember what they sold for way back?
If you’re into analyzing obscure media formats, he has tons of them in his back catalog.