Make it so you can freely download TTS models on your computer and the assistant could use that. Possibly with some first class integration with HuggingFace. This would be great for image generation, general language, embedding, and any other modality for models.