I figured it out myself.
All voice files are under sd/enus.
To find the id of a voice line, search for the line in "TalkMsg" in DS Map Studio, Text Param panel. Only 9 digits "talks" xxxoooooo are voice.
The voice xxxoooooo is reference from "sd/enus/vcxxx.bnk". Load the bnk to Wwise-tooling and search for "Play_vxxxoooooo".
Follow the tutorial to copy entries from "sd/enus/vcxxx.bnk" to "sd/enus/vcmain.bnk". The structure for voice is different from sound, but the principle is the same. 1) find "Play_vxxxoooooo" and "Stop_vxxxoooooo". 2) find the "external_id: ******". 3) Now there should be an entry with "id: ******", search recursively for its "direct_parent_id" until one entry SHARED by "sd/enus/vcxxx.bnk" and "sd/enus/vcmain.bnk". Copy all entries in the sequence from "sd/enus/vcxxx.bnk" to "sd/enus/vcmain.bnk" right above the SHARED entry. Delete unused "children" items in copied entries and add the last copied entry to the "children" of the SHARED entry.
Load the new "sd/enus/vcmain.bnk" in Mod Engine 2.
Now PlaySE(10000, SoundType.Voice, 318010060); in event editing make my Isshin say "Don't let your emotion stay your blade!".