Speech-to-text and text-to-speech for OpenCode. Record voice prompts with whisper-cpp, hear responses via Piper TTS, with LLM normalization through any OpenAI-compatible endpoint.