Wendigo voice cloning.

Wendigos Voice Cloning (Lethal Company mod)

LC Nail 2

Masked has learned to copy your friends' voices. Can you tell who's real and who's fake?

In order for this mod to work properly, it must be installed in all players.

Here's a video presenting this mod.

Disclaimer.

If you are not using Elevenlabs, this mod loads an external binary (about 350 MB) from my github and locally loads an AI model to generate voice clips of masked people. The external binary is 350 MB and the voice clone model is 1.75 GB.

Privacy.

Your voice is cloned entirely locally at runtime. This means that (1) your audio clones are not stored, cloning occurs at runtime, and (2) all processing occurs locally on your machine. It is not externally transmitted. Only generated audio clips are transmitted between players.

When you use Elevenlabs, everything is stored in your account.

Initial Configuration

If you're using Elevenlabs, pre-clone the votes of all players.

Enable mod in Wendigos.cfg. Optionally, enable Elevenlabs and add the API key and voice ID. Also, be sure to set the language and add custom voice replicas (see the "Bonus Features" section).

(Only for local AI models - Elevenlabs users are already disabled.) When you first start the game, you will be prompted to record multiple voice replicas. The currently selected microphone is displayed.

If the microphone you selected is not the one you want to use, click Turn off, set the microphone in Settings, and restart the game.

Governance

'R' starts recording.

'Q' stops recording.

'N' displays the following sound.

If your working language is not English, say what you want in this step and then press 'Q'.

When you stop recording or leave the list of voiced strings, the mode will start generating the voiced strings.

This step can take a long time. When the generation is complete, the game will notify you.

The first time this happens, the mod will load the audio duplication model (1.75 GB) into the Wendigos MOD folder. Since this model is already loaded, subsequent line generation will be much faster.

If you make a mistake and need to record the audio line again, quit the game and set "Record new player audio sample." to true in the Thunderstone "BepInEx/config/Wendigos.cfg" configuration.

Bonus opportunities

Custom voice string.

If you don't like the default voice line. You can customize what Masked can say in different categories of behavior by editing the following file:

BepInEx/config/Wendigos/player_sentences/player0_chasing_sentences.txt

BepInEx/config/Wendigos/player_sentences/player0_idle_sentences.txt

BepInEx/config/Wendigos/player_sentences/player0_nearby_sentences.txt

BepInEx/config/Wendigos/player_sentences/player0_damaged_sentences.txt

New sentences are separated by a new line; you can force AI to say anything (yes, anything).

Each player can have their own voice dialog.

Mask enhancements.

This mod removes the mask and zombie hands from the mask, allowing it to better fool the player. The player's clothes will also be emulated.

Eleven Labs.

Players can use Elevenlabs to clone their voice. This gives much better results and can be more deceptive for the mask. To use this feature, all client voices must already be cloned by Elevenlabs. Everyone can use the same api key, but each player needs a unique voice identifier.

Reaction in real time with Azure and ChatGPT (experimental)

Players wearing masks can respond to the expressions of players in real time. This feature does not support push-to-talk.

AZURE.

Get the AZURE API key from [0]. Create an account and create a new resource group. You will be prompted to create a subscription, but you can choose a free subscription instead of a paid subscription. Try to choose the zone closest to your actual location. The closer you are, the faster the mode recognizes your words and the faster the overall reaction will be.

Then go to the new resource group, click New, look for Azure AI Services, click New, and enter all the information.

Then go to the new Azure AI service and press Keys and Endpoint, enter KEY 1 in the wendigos.cfg file for AZURE API KEY and Location/Region in Region.

ChatGPT

0] and create an account if you don't have one. Go to Profile -> (Company) Billing -> Add Payment Method.

Then go back to Profile -> User API Keys -> View Project API Keys (if necessary), create a new project, create a new key with Project API Key, save it, and place it in ChatGPT API Key wendigos.cfg.

Configuration settings.

Configure all API keys if you have not already done so.

Set General -> Enable mod. and Experimental -> Realtime Responses to true.

Add your name to your Your Name preference if you want AI to recognize who you are.

Optimize Elevenlabs for speed: Speech generation speed was increased, but at the cost of losing most of the emotion in the voice. In general, there will be less stylized and emotional speech.

Speech Probability: The probability that Musk will speak when no players are around.

Possible problems.

Voice lines are out of sync.

Players hear different voices (rarely, but can occur due to lag jumps).

Writing out of buffer results in unity errors.

There is not enough space on your computer to store models and sound lines.

Each player stores voice clone models (1.75 GB), sound samples (~10 MB), and audio tracks (<500 KB each) locally.

FREQUENTLY ASKED QUESTIONS.

What languages does this mod support?

This mod uses XTTSv2 and supports 17 languages: English (en), Spanish (es), French (fr), German (de), Italian (it), Portuguese (pt), Polish (pl), Turkish (en), Russian (ru), Dutch (nl), Czech (cs), Arabic (ar), Chinese (zh-cn), Japanese (ja), Hungarian (hu), Korean (hko), Hindi (hi).

Eleven Labs supports 29 languages. For more information, see https://elevenlabs.io/languages.  

Can I use it with Mirage?

Not yet tested, but probably not (at least when audio is enabled).

Why does it take so long to create audio clips?

The first time you run a mod, you must download the ai model (1.75 GB) and the ai script (350 MB). These may take a long time to download depending on your internet speed. This is done only once and should be faster in subsequent generations. The scripts also load ai models and create audio files for all sound lines. Depending on your computer's capabilities, this may take a while.

Will audio strings be automatically translated into my language?

No, you'll need to enter a sound string into a text file in your language.

Does my voice's clone keep the mod?

No, sound clones are created during the game and are not saved. Only the audio file is saved.

When does mod listen to my voice? If real-time reaction is on, the mod will only hear my voice during the tour. If you have disabled real-time feedback, the mod won't listen to you.

TODO.

[x] Added damage line category.

[x] Masks must play certain categories of lines (such as stalking or inaction) depending on what they do.

[x] Allow players to use ElevenLabs API keys for better voice cloning.

[x] The AI will create new voice replicas between rounds.

[Allow voice cloning of any enemy.

Credits

https://github.com/coqui-ai/tts

Skinwalkers mod by RugbugRedfern

@Kalthun and @notgarrett for their help in testing this fashion.

The Lethal Company Modding Discord

MadLike on Discord for helping with fashion testing and writing real-time guides.