verbiage-gpt is a quick script to enable a console-style interface for relaying the following flow:
- Speak some surrounding flavor into your configured microphone for the script to process.
- Script extracts text from your speech and passes that to ChatGPT.
- ChatGPT generates a response from your flavor plus added character-specific context [configured beforehand.]
- The ChatGPT response is sent to Elevenlabs for audio processing using an AI voice.
- The vocalized response from Elevenlabs is relayed back & played locally.
This is something that can be used to toy around with personally or even use in a wider setting, such as with friends or streaming via Twitch/YouTube.
(This utility essentially mimics DougDoug's setup for doing the same thing on some of his livestreams during gameplay.)
For example:
- You've decided to play Skyrim with a modded Geralt follower.
- You can configure a mock-Geralt voice in Elevenlabs by using personal-use copycat samples.
- You provide the necessary keys & setup values in your own
config.jsonspun out from theconfig_template.jsonreference in the repo. - You run the script and are then able to ask "Geralt" questions via voice, such as where he'd like to go next.
- The mock voice speaking ChatGPT's response is relayed back, allowing Geralt to respond back to you with where he'd like to head to next in Skyrim.
Just clone the project here locally & install the required libraries using pip install -r requirements.txt. Then, configure a config.json file filled out with:
- A working Open AI API key.
- A working Elevenlabs API key.
- Add the name of the voice configured in Elevenlabs.
- Add context info for the character you're interacting with.
python main.py -f config.jsonNote: keyboard for using a key to trigger the microphone recording & cut-off seems to need admin access, so running as admin / via sudo is necessary.
- Store some added context locally or via Pinecone so that a proper memory with the GPT agent is established.
- More to come...