Can anybody please shine some lights to me as I’m fairly new to this all audio fingerprint world.
So for fun I’ve build an engine to scan audio chunks (nothing new of course)
It is working fairly good and fast however the services I used eg ACRCloud and Audd I find them quite expensive.
So came across AcoustID and did manage to convert chunks into fingerprint with Chromaprint etc. Not matter my effort at fine tuning those I will always get a 200 no result reponse.
Reading more in reddit blogs it kind of looks like AcoustID wasn’t design for my use case so that made me feel good as at least I can be in peace with myself again.
However few question is still there, what can you actually do with the lookup by fingerprint ?
what do people use it for exactly, is that actually a way to make it work somehow with what I’m trying to do ?
Tried to contact the acoustid support few times but never got an answer, quite lost in here
also have a paid plan
AcoustID is AFAICT only meant for full files, not chunks, so yes, that’s unlikely to work. The use case we have in MusicBrainz is helping people tag their own music by scanning the files and matching them to metadata.
OK but I scan DJ mixes and break them into 3 min chunks should that resemble a full file track of course if this is not overlap/mixed with an other track ?
The point is: Another user needs to report the almost identical track to the AcoustID website first.
Only then later users can “compare” their own songs with those already submitted.
The chance that some other user has already submitted the same “scanned 3 min DJ mix chunks” is very low IMHO.
If you have the complete song ina file, e.g. from a download or CD rip, then you also have the length.
The AcoustID server was designed for this full file support, with the use case of audio file taggers like MusicBrainz Picard in mind. The lookup uses a fingerprint from the start of the audio file (first 120 seconds) and the total length to find matches.
If you have multiple tracks in one file, nicely one after another without fading, and you cut at roughly the correct position (there is some threshold of a few seconds shift that would be allowed), then yes, the fingerprints should be usable. But of course it also requires the track in question to have submitted fingerprints before that are reasonable similar.
What the AcoustID server is not suited for is to take an arbitrary audio snippet from a file and do a lookup.
I can’t find it right now, but luks once wrote about this design decision. Theoretically a database would be possible that allows lookups of such audio snippets, but at much higher hosting requirements. Limiting the design to the use case of full track lookups allowed him to actually operate this service.
Your use case seems to be “identify this music a DJ has just played”.
As I understand it, AcoustID aim is to “identify this whole track I have ripped”. I have an MP3\FLAC file ripped from a CD and want to know what it is. Think of someone who has a hard drive filled with music from dodgy downloads like Napster, eDonkey, Torrents, iTunes. Or a heap of files ripped from their own or mate’s CDs and badly tagged. It is these whole data files that are being fingerprinted and recognised. Mainly due to them being neatly split into the musical tracks in a consistent way.
If you are scanning a DJ mix then it will be hard to catch the actual start of the file. Especially if it is mixed into the track that came before it. A mixed track creates a different fingerprint and will be hard\impossible to identify with AcoustID.
I don’t think AcoustID is really going to be able to help your use case.
