Are there any remotely comparable open source models? Fully multimodal, audio-to-audio?