Anyone knows if there's a current sota or benchmark to know what the top voice-to-voice LM is ? By this I mean you talk to it in voice, and it responds in voice (natively, not the cascade tts/stt pipeline)
