Some new sampling. It could have been anything, but I like Journey and this one is from 1983, so it fits
Needs 16kB RAM expansion. The main routine is 206 bytes, the rest is the highly compressed data. Keep in mind a standard mp3 file is typically several MB!!!
I once had a VIC-20 "Speakeasy" speach cartridge that I ordered from an ad in Compute! magzine, which used a Votrax chip. It had somewhere around 64 phonemes, but some of them seemed to be similar enough to each other as to be redundant. The phoneme was selected by poking its number onto a register; the phoneme continued to play until another phoneme number was poked into the register. The value stored in a status register changed to zero then a new phoneme started, then changed to one when the phoneme had played for a certain amount of time that varied for each phoneme value; your program monitored this register to determine when it was time to move on to the next phoneme.
But the phonemes themselves sounded as if they were looped sound samples with a duration of a fairly small fraction of a second. You might be able to squeeze these samples into the VIC's memory with room left for a decent program.
The old Speakeasy cartridge, if I remember correctly, had an RCA connector output that could be sent to any standard amplifier, but there were instructions describing how to solder a wire into the RF modulator to output to the same TV set as the VIC20.