About this mod
Updating for Patch 6 in progress. Look for files with version 1.8.0.
Videos section is updating as I update characters.
This is the first pass. New lines (50-100 per character) will be added in 1.8.1
Many male to female and female to male voice conversions.
- Permissions and credits
- Changelogs
Updates:
Added voices for UI clicks and world interaction!
Added SPELL VOCALS some, but not all, spell voices (Gale, Wyll, Astarion, Shadowheart, Lae'zel, Halsin)
For another solution to spell casting voices see: Non-Verbal Spells that simply removes the vocal component of spell casts but keeps all spell sound effects.
Scope:
These mods provide AI assisted and classical audio engineering voice conversions for Origin characters.
Installation:
Each character voice is a separate file. Select as few or as many as you would like.
Use vortex.
For BG3 mod manager, install the .pak file.
Removal:
These files are completely safe to add and remove at will and will - so long as Larian supports loading saves with changes to the mod list.
Known Issues:
Not all spells are covered. Grunting noises are not changed.
To clear up some confusion:
This mod uses a number convention for voice styles. Every style is unique and hand selected to mesh with the character (in my opinion). So every "1" and "2" is a unique voice. Numbers indicate more unique options for the same character.
Nolester also has an excellent selection of voices included in his mods. His naming convention uses letters. Voice A, B, Etc. Each mod with the same letter has the same AI voice style. So Astarion A sounds like Gale A. (I think)
Background:
Initially I tried to use classic audio engineering techniques with Reaper and Audacity to shift the pitch and formants. I was successful, but the vocal range of the actors is too great to batch process files in that way. It would have required file by file processing and countless hours of work. So, I transitioned to using RVC and voice models then cleaning up the audio with Reaper. It’s much easier to change a song or speech in narrow key than to change 9000 voice lines that range from happy and pitchy to soft and intimate. The vocal range of the actors for BG3 was significant. Avoiding all artifacts, even with AI, would require a file-by-file adjustment to fine parameters. This solution is good 95% of the time. 5% of the time it will be noticeable.
Other notes:
Nolester (Nolester's profile) has links to AI voice mod resources.
I suggest using appearance edit mod by Eralyne along with a body mod by snk4211.
Gale:
Astarion:
Wyll:
Shadowheart:
More Videos in the Videos section.
Additional Process Details:(This section is for other modders, or the curious.)
![](http://static.tvtropes.org/pmwiki/pub/images/6db_3.jpg)
For core dialogue lines, dialogue sounds files have a UUID number that corresponds to specific characters. This makes it very easy to deterministically select all dialogue files for a specific character. Previous voice mods, and these voice mods, take advantage of this naming convention.However, other sounds such as character comments in the world, spell casts, and combat noises are located separately and do not use this naming convention. For these sounds the file names use unique numbers that cannot be tied to characters. The modding community would require an index for characters that lists sound files. So far, I have been unable to find game files for spells, combat noise, or characters that link these numbers. But I was able to reverse engineer a process to make some progress:
There are sound banks. These .bkn files contain group references for sound ids. (sounds have a sound id that matches file names. Sounds have a parentID. Likewise sounds parents have parents. This is how they can be grouped.) They can be grouped up to 6 levels. I had hoped this would group all the way up to the character level, but it instead groups to sound types. For Voices, the 2nd level group is character specific for a sound type. For example: all of Astarion's lock-pick voice lines. The use of some Python script wizardry allowed me to reduce the number of files from ~100000 to 700 groups. I manually identified the voices by listening to clips in the 700 groups. This was not a particularly easy task as some groups are stealth whispers that are hard to identify.
This solves the problem for voices. Unfortunately, it doesn’t work as well for spells and combat noises. Which are frankly very hard to identify by ear and have a larger number of groups. For modders: This initial process is brutal and requires python and xml parsing skills. The “best” way to get the list of sound names would be to extract the list of names from the mods I upload. Then use a python script to copy the sounds with those names into a working directory.
I made significant progress indexing the spell casting voices. I can get large chunks of spells identified. Unfortunately, there are too many isolated spell groups for each character (1000's). So even if I change 50% of the spell voices in a single batch for a character, I will be missing many. Additionally, the AI voice change quality is atrocious. Spell casting effects have so much reverb and distortion built in that the AI models simply break. Update: by doing more aggressive pre-processing I was able to make this passable, but not perfect. If anyone want's to help, there is a file in the miscellaneous section that contains the spell groups in need of identification.