Neural audio codecs: how to get audio into LLMs