Show HN: Caramel: a speech-to-speech conversation model trained only on audio4 points by naklechaRead the articleDiscussion on Hacker News