It uses the OpenAI text-to-speech API, and Vosk Speech Recognition Toolkit.