I’d think of it as a caching thing.
You have an audio resource at some URL. It’s just PCM bytes, no metadata. A wav file would match this description. But I’m thinking about next-generation audio formats that have real compression, are accessible to CORS, and can be fed into web audio APIs.
For the most part the file would be kept with its URL. The bytes returned by the URL and the URL itself wouldn’t be decoupled, so that one could travel without the other. Given that the client has the URL and the URL is enough to get a metadata file, the metadata and audio do travel together.
The concept for metadata in this post is a sibling to the web audio API. It’s a web metadata API.