[Solved] Voice files extracted from the game for Voice Cloning (AI-generated new dialogues)

[Vel (Tien-Yun) Wu]

Hello,

I currently am doing an experiment of building an expanded universe where existing characters will be involved in excitingly new scenes/dialogues/conversations that the original publisher of the game did not intend.

The idea is to create new stories for old characters, where you have characters like Jaina, Kael and Thrall saying things the original voice actors never said themselves.

Because the group of people I belong to obviously don't have the money to hire the very capable and expensive original voice actors, my personal mission is to use this opportunity to investigate the feasibility of using voice cloning technologies to create naturalistic, new voice lines for the new dialogues that are written.

The idea is essentially using artificial intelligence to which I will feed voices of the original voice actors/actresses so they learn the patterns of audio data. After this training is done, I will feed the AIs some texts (these are newly written dialogues by creators) and they will be able to produce voices for those, without the original voice actors' participation.

I will eventually report the results of this investigation to my peers to help deciding whether this is a feasible solution for the project(s) we might each be involved in.

Technologies I will test this idea out with are (more might be added in the future):

1. Corentin Jemine's Real-Time-Voice-Cloning solution:

- Demo:

- Source codes:
CorentinJ/Real-Time-Voice-Cloning

2. The Respeecher solution made by this fine company from Urkraine:
Voice Cloning Software for Content Creators | Respeecher

3. Resemble AI
AI Generated Voices ~ Resemble AI


Therefore, I need voice files extracted from the game. This can be in flac, wav, mp3 or any format that
makes sense for an audio.

Specifically, voice files of lines spoken by Arthas, Tyrande, Thrall, Maiev, or any character that is a protagonist would be best. This is because they would provide the greatest number of samples for machine learning purposes. If anyone here has some on hand, or know how to extract them, I would be deeply grateful if you can reply to this thread of message me in private.

Thank you in advance.

 

[Vel (Tien-Yun) Wu]

Replying to myself. It seems these audio files can be obtained one by one using the Sound Editor module in the World Editor.

I think this request can be assumed as solved now, unless anyone knows of a way to extract these files all at once. One by one is a bit inefficient, but it works, and that's fine

 

[Teitan]

I think you can (i might be wrong) access all at once with an MPQ browser, which is a program that lets you open (and extract) the game data files - outside of the world editor

EDIT: here is an old one, i am unsure if it works with the current version of the game, most likely not, but it's not particularly hard to get the old game files, Ahoy!

 

[Footman16]

I think you can (i might be wrong) access all at once with an MPQ browser, which is a program that lets you open (and extract) the game data files - outside of the world editor

EDIT: here is an old one, i am unsure if it works with the current version of the game, most likely not, but it's not particularly hard to get the old game files, Ahoy!

For the most recent version you'll need lasiq's CasCViewer, however if you still have the old MPQs like I do then you can still use the MPQ browser.

 

[Kakerate]

Post an update! which unit sounds have you used to synthesize new lines for?

 

[Vel (Tien-Yun) Wu]

Post an update! which unit sounds have you used to synthesize new lines for?

I appreciate your interest in this.

So...in short, not many sounds at the moment. You can find the very primitive results here in my pastebin:

LINK to pastebin

The three voices that I've attempted to synthesize are human Arthas, Jaina and Tyrande's.
As far as training data goes, the first two's voices are taken from Warcraft 3, whereas the last one is taken from Heroes of the Storm (this part was courtesy of another Hive member)

This is also where I'm currently stuck at: unable to make the synthesis better.