zlacker

Azure Speech tts is capable of doing this with SSML. I wouldn't be surprised if it's what OpenAI is using on the backend.