It's almost certainly Google SoundStorm, a traditional TTS trained on dialogs from last year: https://x.com/jonathanfly/status/1675987073893904386