I have the weirdest issue, but I'll start with the specs:

-Neoware CA2 Thin Client
-VIA C3 Nehemiah+ CPU @800 MHz
-256 MB SDR-133
-S3 ProSavage IGP (32 MB shared memory)
-2x 32GB CF cards via IDE
-Labway LWHA111800 sound card with ESS Audiodrive ES1868f(ESFM) and Admos QDSP700 wavetable (General MIDI)
-Windows 98 SE
-Virtual CD ROM drive via Daemon Tools
-Might & Magic 4&5: World of Xeen game

I set the game's sound configuration to SB Pro for both "Music" (FM synthesis) and "Digitzed Speech" (which in this case only refers to ingame shops and the sound effects of monsters).
However, on the 2 CD version of the game, every NPC, no matter how unimportant, has full voiceover, probably done by five different interns, I don't know). This is always done via CD audio, no matter what you set the two sound options to.
I use .bin/cue images of my two original game discs, and when playing via DOSBox on my Windows 10 system, with the -ioctl parameter for mounting the images,I hear music, sound effects, and speech.
Now on my Neoware CA2 Thin Client with a real ISA sound card, I also hear music and sound effects in the game, but when an NPC would talk, instead of hearing speech, the whole system is slowed down until the voice track would be over.
However, using Windows 98's CD player, I can choose all the audio tracks on the (virtual) CD and listen to them perfectly fine.
This makes no sense to me, as the game runs in the MS DOS window that uses the same drivers as everything else I start from Windows. If it wouldn't, I wouldn't hear sound at all and my (virtual) CD ROM drive would not be accessible at all.
So in theory, it should have the same access to the audio tracks that works perfectly fine on the same machine for the CD player software. But in practice, no voices.

I know it's a long shot, with the game being so old, so niche and the problem being so weird, I doubt this can be solved. But I at least wanted to try reaching out, since I cannot imagine it being impossible to solve, when the CD player software actually does play the very same voice tracks that WoX can't for some reason, on the same machine with the same OS.