What do you think about the recent SAM audio model by meta? https://ai.meta.com/blog/sam-audio/