Why is the voice 'onyx' available on audio-preview but not available on realtime-preview model

I tried audio-preview api and it works for voice: onyx. But realtime api doesn’t accept it. I thought both APIs used the same underlying models, so wondering why it works on one but not on other.