I recently listened to some "tracks" of Mitokoumon NSF, and I've heard some creepy voices in it.
Listen at indexes 0--4, 18--24, there's something like some words in Japanese (and even a creepy laugh), but it isn't using DPCM samples, it is rather updating $4011 directly, or using very very short samples―I don't know which.
I recall seeing some
videos of some Famicom adventure games that had this recently. I had thought it was the FDS output, but it
sounds like a pack of very finite samples strung together to make words...
That's not ESS, is it? "Another visitor. Stay a while. Stay forever!"
(
article and
Flash-based audio player)
I guess it might be possible to use the FDS wave RAM to implement ESS playback with less CPU attention than a timed-$4011 solution.
It is in fact very similar in sonority, it seems however that ESS outputs voice at a lower sampling rate.
A local game store here has a copy of the game for $20.
It's definitely doing it via $4011 writes. There are no DPCM samples being played.
They do have a strange quality to them, though, almost like the original recordings were analyzed and then played back through a speech synthezier. Like, it feels like pitch of the voice updates at a regular interval, rather than continuously.