It’s not directly usable for this, because it was not designed for it. AcoustID, the actual server side lookup service, was created to identify whole audio file, and intentionally holds only fingerprint for the first 60 seconds of a song. I can’t find the source right now, but Lukas Lalinsky, who created and operates AcoustID, once mentioned that this restriction even made it possible for him to operate this on an affordable basis, because it limits the required data storage and processing power.
The other part is Chromaprint, the fingerprinting algorithm. It also was designed to identify clean audio, and it does not work well with very short snippets.
The following mailing list discussions have some details about this:
https://groups.google.com/g/acoustid/c/OQ4332mbhF0/m/L8XHAMaBuI8J
https://groups.google.com/g/acoustid/c/C3EHIkZVpZI/m/8jDpwEyQk9oJ
I think the issues are probably not so much legal ones. A database of fingerprints likely has no legal problems. There might be some patent issues, but that’s something we have in other parts of FOSS software as well.
The technical side should not be underestimated. But as indicated by the answer from the FSF there are existing open source audio fingerprinting solutions, and there is extensive audio metadata on e.g. MusicBrainz. I’m sure someone could come up tying this together for a FOSS audio lookup server.
But the biggest challenge then probably is to operate such a service and gathering the data (fingerprints). Full audio fingerprinting surely uses significantly more storage compared to what AcoustID has right now.