but since most of their new data is directly from labels, I don't expect something like this is their interest
2007-11-04 30841, 2007
brianfreud
but e're not talking, I think, an alternate to mbids. We're talking a secondary db of fingerprints to the same tracks that have puids - the last.fm fingerprints may be otherwise meaningless, but they allow better identification of semi-useful puids which are incorrectly attached
2007-11-04 30811, 2007
luks
brianfreud, ther fingerprints on it's own identify nothing
2007-11-04 30827, 2007
luks
you need a way of matching them, which is not simple A == B
2007-11-04 30823, 2007
luks
you can tell if two puids are them same, because they are only IDs
2007-11-04 30832, 2007
luks
but matching for example libofa fingerprints is more tricky
2007-11-04 30835, 2007
brianfreud
If my local library uses their own numbering system, it may be meaningless outside that library. But if I have their numbers, and I have LOC library numbers, assuming the local library's numbers are well attached to each book, I can use the 2 sets of numbers to cross-verify the LOC numbers, to try and spot LOC numbers incorrectly assigned to various books.
2007-11-04 30806, 2007
luks
brianfreud, fingerprints are not numbers
2007-11-04 30818, 2007
luks
fingerprints are a few kB of data
2007-11-04 30824, 2007
brianfreud
but but they are unique identifiers
2007-11-04 30829, 2007
luks
no, they aren't
2007-11-04 30851, 2007
clever
could treat it like a crc hash maybe
2007-11-04 30852, 2007
luks
two different fingerprints can resolve to the same PUID or TRM
2007-11-04 30855, 2007
brianfreud
ok, yes, one fingerprint can resolve to multiple tracks (which I finally proved for puids yesterday, lol
2007-11-04 30825, 2007
luks
clever, that would make it as useful as crc hash == useless :)
2007-11-04 30840, 2007
clever
crc on mp3 wont allways get a proper match
2007-11-04 30855, 2007
clever
since a slight typo in the song title(in the id3 tag) can make the crc not match
2007-11-04 30809, 2007
clever
also 2 diff encodes of the same song will cause slightly diff compressed data sets(which wont match)
2007-11-04 30833, 2007
brianfreud
yes, but we're not talking CRC, we're talking accoustic fingerprints
2007-11-04 30811, 2007
luks
the point is that accoustic fingerprints are not identifiers on their own
2007-11-04 30820, 2007
clever
which is just like a crc of the way it sounds(not the raw data)
2007-11-04 30822, 2007
luks
they are sets of features extracted from the audio
2007-11-04 30824, 2007
brianfreud
sure, I agree
2007-11-04 30844, 2007
luks
and you need a method to actually make the extracted data useful, and resolve them to IDs
2007-11-04 30847, 2007
clever
but 2 things which have the same id the prog may think 'sound' the same
2007-11-04 30850, 2007
brianfreud
but I think there's some assumption here that puids are only attached by users to the correct tracks for those puids
2007-11-04 30800, 2007
luks
which is the most tricky part and I haven't seen open source software to do that yet
2007-11-04 30832, 2007
luks
brianfreud, I'm talking about fingerprint->puid matches, which is done completely by musicip closed servers
2007-11-04 30854, 2007
brianfreud
ok, I'm talking puid-->mbid, not fingerprint-->puid
2007-11-04 30828, 2007
brianfreud
fingerprint--> puid, we assume, is always correctly done by the MIP servers
2007-11-04 30840, 2007
brianfreud
puid-->mbid is the part where error creaps in
2007-11-04 30822, 2007
luks
yes, but in case of last.fm fingerprints there is no public way to do fingerprint -> some ID
2007-11-04 30834, 2007
luks
which makes them useless for anybody except last.fm
2007-11-04 30839, 2007
HairMetalAddict
"We are frantically working on the server architecture in order to get a public-facing lookup service ready as soon as possible." - from blog.last.fm initial announement. Nothing about OS'ing the actual code that I can see, just a lookup service.
2007-11-04 30805, 2007
HairMetalAddict
RJ also mentions connecting the fingerprints to MBIDs as well.
2007-11-04 30823, 2007
brianfreud
I'm saying, given several mbids and several puids, and none of the source files that generate the fingerprints that resolve to the puids, having anything else - even if it's just somoene else's fingerprinting scheme's raw #'s - to compare against gives you some way to possibly ID which puids are attached to the wrong mbids in a small set of cross-linked puid-mbid pairings
2007-11-04 30850, 2007
luks
brianfreud, no, it doesn't you anything
2007-11-04 30812, 2007
luks
for example: you have fingerprint sfhgjhjhsdgfjhfgsdlfkjhdslkfhs234fkjhweiyewuri345y4 and kejfhgjherkjthertuiyrewieuyrewiyrue345435
2007-11-04 30817, 2007
luks
are they the same?
2007-11-04 30826, 2007
brianfreud
sure, who knows?
2007-11-04 30829, 2007
luks
or, resolve theyt to the same track?
2007-11-04 30849, 2007
luks
this is that makes them useless for cross-checking PUIDs
2007-11-04 30829, 2007
brianfreud
but say I have mbid1 with puid1 and puid3. mbid2 with puid2 and puid3. mbid3 with puid1 and puid2. Theoretically, given multiple people also fingerprinting the exact same mp3s, I could also get lkjglglhjglglkg on mbid1 and mbid2, and 1234dfjhgkhkgfkxjgjh on mbid2 and mbid3. I might also have 20 unuseful fingerprints accross the 3 files - but those matching ones could be helpful in cleaning up the puid assignments to the mbid
2007-11-04 30824, 2007
luks
you will almost never have identical fingerprints
2007-11-04 30835, 2007
brianfreud
(in this case, leading to a strong suspicion that since 1 ~~~ 2 and 2 ~~~ 3, 1 most likely == 3)
2007-11-04 30856, 2007
luks
any encoding of the same track will have a bit different fingerprint
2007-11-04 30816, 2007
brianfreud
luks, from different sources no... but umm, many people often have bit for bit identical tracks, as they all, umm, source from the same rip...
2007-11-04 30829, 2007
luks
that's no more usefull than crc then
2007-11-04 30831, 2007
brianfreud
:D
2007-11-04 30801, 2007
brianfreud
well, sure, you could replace last.fm fingerprints anywhere above in my argument with crcs, and the same would still hold true, I agree
2007-11-04 30847, 2007
brianfreud
I'm not saying a raw fingerprint, or a crc, is all that helpful - just that in certain specific situations it *could* be useful... rarely
2007-11-04 30809, 2007
Amblin joined the channel
2007-11-04 30842, 2007
brianfreud
99.9% of that data would be junk... it'd be the .1%, or .0001% which was in there 2 or more times which could be useful
2007-11-04 30809, 2007
HairMetalAddict
"We want to work together more closely with MusicBrainz and maybe at some point MusicBrainz might even want to switch to the fingerprinter we’re using? It really does a pretty good job and the source code is already out there and the web service will be open." - staff member e1i45 in a blog.last.fm response, heh
2007-11-04 30813, 2007
brianfreud has quit
2007-11-04 30824, 2007
brianfreud joined the channel
2007-11-04 30827, 2007
metafollic has quit
2007-11-04 30820, 2007
metafollic joined the channel
2007-11-04 30827, 2007
mudcrow joined the channel
2007-11-04 30820, 2007
metafollic has quit
2007-11-04 30818, 2007
metafollic joined the channel
2007-11-04 30843, 2007
brianfreud
"Please consider importing the artist/release/track(s) you are looking for from freedb."
2007-11-04 30800, 2007
brianfreud
How exactly does one import an artist from freedb? :D
2007-11-04 30829, 2007
xlotlu
is the site getting funny.. or my isp?
2007-11-04 30822, 2007
brianfreud
I think it just died
2007-11-04 30857, 2007
xlotlu
well bah'
2007-11-04 30809, 2007
xlotlu
i was in the middle of fixing some buxtehude mess
2007-11-04 30826, 2007
LotR
we could try getting access back by doing a seance here
2007-11-04 30830, 2007
brianfreud
keep the page open - you ought to be able to go back one page and keep going when the site is back
.jpg){kind=link}