The Masked have learned how to copy the voices of your friends. Can you tell who's real and who's fake?
This mod requires every player to have the mod to function corrcectly.
Here's a video I made showing the mod off!
If you are not using Elevenlabs, this mod downloads external binaries (about 350MB in size) from my github and loads an AI model locally to generate voice clips for the masked. The external binary is 350MB, and the voice cloning model is 1.75GB.
The clone of your voice is done at runtime entirely locally. This means that (1) there is no stored clone of your voice, the cloning is done at runtime, and (2) all the processing is done locally on your machine. It is not shared externally anywhere. Only the generated voice clips are shared between players.
If using Elevenlabs, everything is stored on your account.
Wendigos.cfg. Optionally enable Elevenlabs and add API key and voice ID. Also be sure to set your language and add custom voice lines (see Bonus Features).If the selected mic is not the one you would like to use, simply click "close", set the correct mic in settings, and relaunch the game.
Controls:
If your preferred language is not English, say anything you'd like during this step, then press Q. Try to record at least 1 minute of audio.
Once you stop the recording or finish the list of voice lines, the mod will start generating your voice lines.
If you make a mistake and need to record your voice lines again, exit the game and set "Record new player sample audio?" to true in "BepInEx/config/Wendigos.cfg" in your Thunderstone config.
Not a fan of the default voice lines? Customize what the Masked can say in different behavior categories by editing the following files:
Separate new sentences with new lines. You can make the AI say anything (yes, anything).
Each player can have their own unique set of voice lines to better suit things that they might say.
This mod removes the masked masks and zombie arms to better fool players. Player clothing is also mimicked.
Players can use Elevenlabs for voice cloning. This produces far better results and makes the masked much more deceptive. For this to work, every client must have their voice already cloned by Elevenlabs. You can all use the same api key, but each player needs a unique voice id.
Allows the masked to reply to things players say in real time. This feature does NOT respect push-to-talk.
go to https://portal.azure.com/ to get your AZURE API KEY. Create an account, from there create a new resource group. You will most likely be prompted to create a subscription, when you're doing that it's fine to pick the FREE subscription instead of pay-as-you-go. Make sure to pick a region that is CLOSEST to your real-life location, the closer it is the faster the mod will recognize what you're saying and make the responses overall quicker.
After that's over, go into your new resource group and press create, search for "Azure AI services", click create, punch in all the data, again put the region that's closest to you.
After that's done go into your new Azure AI Service and press on Keys and Endpoint. Put your KEY 1 into wendigos.cfg in AZURE API KEY and put your Location/Region in Region.
go to https://platform.openai.com/docs/guides/text-generation and create an account if you don't have one. Go into your profile -> (under organization) Billing -> add a payment method
After which go back to Your profle -> User API keys -> View project API keys if it requires you to do so, create a new project, in the project API keys create a new key, save it and put it into wendigos.cfg in ChatGPT API key.
Set all api keys if you haven't already.
Set General -> Enable mod? and Experimental -> Realtime Responses to true.
Add your name to the Your Name setting if you'd like the AI to know who is who.
Optimize Elevenlabs for Speed: speeds up voice generation at the cost of losing most emotion in the voice. Less stylistic and emotional speaking overall.
Talk Probability: how likely the Masked is to talk when no players are nearby.
Writing past end of buffer unity error (you have a generated line that is too large)What languages does this mod support?
This mod uses XTTSv2, which supports 17 languages: English (en), Spanish (es), French (fr), German (de), Italian (it), Portuguese (pt), Polish (pl), Turkish (tr), Russian (ru), Dutch (nl), Czech (cs), Arabic (ar), Chinese (zh-cn), Japanese (ja), Hungarian (hu), Korean (ko) Hindi (hi).
Elevenlabs supports 29 languages. Find out more here: https://elevenlabs.io/languages
Can I use this with Mirage?
I haven't tested it yet, but probably not (at least not with the voice features enabled).
Why is the audio clip generation taking so long?
The first time the mod is run, it needs to download the ai model (1.75GB) and ai script (350MB). These can take a while to download depending on your internet speed. This is only done once, so it should be faster on future generations. The script also loads the ai model and generates audio files for all the voice lines. Depending on your pc specs, this can take a while.
Are the voice lines auto-translated into my language?
No, you'll need to write your voice lines in your language in the text files.
Does this mod store a clone of my voice?
No, the voice cloning is done at runtime and is not saved. Only the audio files of the voice lines are saved.
When is the mod listening to me? If you have realtime responses enabled, the mod listens only during rounds. If realtime responses is disabled, the mod is never listening to you.
