How can I add variability to TTS in Azure AI Speech?

I'm using the en-US-NovaTurboMultilingualNeural voice in Azure AI Speech's text-to-speech (TTS) service.

When I've used other TTS services (e.g. OpenAI, ElevenLabs) each generation of speech results in a slightly different reading. ElevenLabs even has a seed parameter to control this variability in behaviour.

I realise that I can use SSML to customise styles and stress etc, but is there any way (e.g. a seed parameter) to automatically try a different reading of my text input? Or is this TTS service designed to be completely reproducible and controllable?