As someone who’s had some incidents with DSP code, the end of the recording sounds like it may be playing some part of memory that isn’t an audio buffer. I wonder if there’s actually a “DTMF injection” possibility here…
You can hear the windows XP message box sound right before that. Which surprises in two ways: a) they're still using windows XP (ok well we still do too at work for some appliance from the power company). b) it seems you're not hooked into the machine via some modem or virtual-something over lan, but something that connects to the sound card, otherwise I've no idea how system sounds that always play on the default card would end up in the phone call. That means there's one machine handling one call at a time.
didn’t have much luck with multiple baud rates and modulations on minimodem, no discernible ascii, but someone might have more luck looking at the binary output
It wouldn't be encoded in a modem protocol. If that's indeed binary data, then most likely we're hearing binary data interpreted as being PCM wave data.
> shared it with a few people who Know Telephone before I posted it here, and their theory is that what we're hearing at the end is the audio path going open-circuit when the PC crashed.
It probably blue-screened, and we're hearing the EM interference from the CPU or I/O controller hub as Windows writes a minidump, then begins waiting for a debugger to attach (the blerps at the end being scans for connected serial port, PCI, network or 1394 debug hardware)