Thanks Lars, yes, I did that.
I've now did it again from scratch, documenting every step:
1. Started with a clean copy of Freddy CD
2. Moved all AUDIO/ files to the parent directory
3. Verified that it's OK in ScummVM. (no warnings, voices are OK).
4. Opened the game in Kawa's latest SCICompanion (3.2.3.0)
5. Opened Message.230
6. Exported "Noun 4, Verb 1" (the response to "look barrel") and "Noun 28" (Willy's speech) to .wav files
7. Imported them, swapped (such that we should hear about rain barrels when the game begins, and 'look barrel' will give Willy's speech in response)
8. Verified in SCICompanion that each message has the opposite voice
9. Saved with ctrl-s
10. Clicked Tools->Repackage audio
11. Waited a little...
12. Opened in ScummVM, got a warning upon start:
WARNING: Failed to find volume for map.540!
13. Instead of Willy's speech, indeed I heard the rain barrel answer - which is great.
14. However, when I "look barrel", there in no voice, and in the ScummVM's console there is:
WARNING: Resource type mismatch loading audio36.230(4, 1, 0, 1)!
WARNING: resMan: Failed to read audio36.230(4, 1, 0, 1)!
WARNING: Failed to find audio entry (230, 4, 1, 0, 1)!
WARNING: wPlayAudio: unable to create stream for audio tuple 67174401, module 230!
Please help...