As those of you who have looked into this file-related matter know, most of the tools available here and on other sites are mainly for model extraction, and voice files cannot be extracted without an update from a knowledgeable person(voice files are in particularly small demand).

However, as an amateur, I was able to find the voice files and listen to bits and pieces of them by comparing them to the music files I had placed with them.
The file I found this time is from Masterwork Apocalypse.
This is the hexadecimal of the .npk before it was determined to be audio.
And this hex had the same description as the other wem files.
After forcibly transcribing the description and changing the format, I was able to play it in foobar 2000, and I knew that the voice was in this file.
Here is the audio played back.
But that was as far as an amateur could go.
I don't know how much of this is one file, or how to cut out each part.
And even a quick look at the file shows "RFF#>. .WAVEfmt" and there are some parts that need to be modified.
Is there a way to write a script to solve these problems?