Offline version of church-of-gpt
npm install
- Install ollama
curl -fsSL https://ollama.com/download/Ollama-darwin.zip -O
ollama run dolphin-mistral
ollama serve
- Install Vosk for the voice recognition: https://alphacephei.com/vosk/install
- Download a model and add to program directory: https://alphacephei.com/vosk/models (see code for the recommended model: currently vosk-model-en-us-0.42-gigaspeech)
- Customize prompt.txt and set correct permissions:
chmod 644 prompt.txt
- Install
sox
brew install sox
- Install
piper-phonemize
git clone https://github.com/rhasspy/piper-phonemize.git
cd piper-phonemize && make
sudo cp install/lib/* /usr/local/lib
- Install
piper
git clone https://github.com/rhasspy/piper.git
cd piper
brew install cmake
make
- Download speech model and add to program directory: Model download link
- test piper
export DYLD_LIBRARY_PATH=$DYLD_LIBRARY_PATH:/usr/local/lib; echo "hello" | ./piper/install/piper --model en_GB-alan-medium.onnx --output_raw | sox -t raw -r 22050 -b 16 -e signed-integer - -d pitch -100
./start.sh
- Done with the fun?
Ctrl + C
- On a brand-spankin new Mac:
- we install homebrew
- then we install node via homebrew
- node was failing to install anything
- because python wasn't installed properly
- so we needed to install python via homebrew
- and then we needed to install node-gyp globally
- and then finally we were able to install everything we needed