Allow voice chat players/audience to control your PC with hot word detection (with minimal CPU usage).
This example makes it so that if someone says "americano
" in voice chat, I catch on fire 🔥 in-game. (American accent required)
They can then extinguish 🧯 me by saying "terminator
".
- False positives are common so this is just a toy.
- No warranty of any kind.
- Something like 30-500MB of extra RAM (
python's fault, sorry
). - The program needs about 1% of CPU power with my old Intel 4th generation i7-4770.
- Python 3.11+ installed.
- VRChat avatar knowledge OR programming skills in python to adapt to another game ( Source engine example "coming soon", based on valvecmd )
- If VRChat:
- An avatar you have uploaded yourself or an existing OSC parameter (see how to use).
- Some knowledge about the OSC system
- If Garry's Mod / TF2:
- TODO
- Download this repository, extract anywhere.
- Run
install.cmd
- Open
%userprofile%\AppData\LocalLow\VRChat\VRChat\OSC
and find your Avatar based on the ID.- Find your avatar ID by opening https://vrchat.com/home/avatars and clicking your avatar and look at URL.
- Find the boolean parameter you want to toggle on/off, in example case:
... { "name": "fire_effect", "input": { "address": "/avatar/parameters/fire_effect", "type": "Bool" }, ...
- Open
main.py
and change the/avatar/parameters/fire_effect
into your own boolean value. - Optionally: Change keyword
- Available keywords (Note: new picovoice that requires a license key has way more keywords to choose from. See keywords here):
- americano
- blueberry
- bumblebee
- grapefruit
- grasshopper
- picovoice
- porcupine
- terminator
- Available keywords (Note: new picovoice that requires a license key has way more keywords to choose from. See keywords here):
- Launch VR BEFORE launching this program (This program uses the primary audio source on launch, which changes when you start VR)
- Launch
run.cmd
or clickmain.py
- Ask someone to say in american accent
americano
or test with the example sound file - Catch on fire or whatever 🤷
Based on old version of https://github.com/Picovoice/porcupine/ that does not require a license key. New version would allow using more keywords.
- Add openWakeWord if CPU usage is not a concern.
- Only record the application instead of music player or similar by using https://github.com/microsoft/Windows-classic-samples/tree/main/Samples/ApplicationLoopback
- Add Source 1 engine support