I don't think that would work, particularly if using Elevenlabs voices. Prompting the voice to speak slowly, even through their native TTS, is a hit-or-miss, and further it requires manually editing out the prompting such as "Hi, how can I help you? - **she says slowly**" (removing "she says slowly") so I'm not sure if it will work simply by prompting the LLM. After all, the voice ultimately is generated from the TTS of the voice provider and not the LLM.