#musicbrainz

Knio_ is now known as Knio

Berkz has joined #musicbrainZ

is it normal for a trm lookup in database to find multiple files

? Why is this?

Imperfect matches.

wheels: can 2 audio files actually have the same trm signature?

I would assume so.

Berkz: But remember -- you're usually working against comparing lossy audio to more lossy audio.

how come they didn't use something like a md5 hash where no 2 files can have the same signature... ?

Berkz: Because then two files ripped on different machines would never match.

Berkz: Like if we both buy a CD and I rip it with LAME and you rip it with iTunes -- even though they're the same song at the same bitrate they wouldn't match.

i c

good point

And of course md5 sums can be identical too. ;-)

That's the magic of TRM -- it's not completely deterministic. It would be pretty useless if it was.

I think they can't be the same size and have the same sum unless it's exactly the same file..

Well, deterministic isn't the right word there, but I think the point was still clear. :-)

Berkz: Who said they had to be the same size?

Berkz: And statistically that's not possible. You can't reduce a gigbyte of data to a small hash without having a number of overlaps in possibilities in that gigbyte.

my point is that md5 hash would be the perfect matchmaker cause then u could be sure if the file is the same size...

Berkz: Otherwise you would have just invented the world's best data compression. ;-)

wheels: what app do you use for you id3 tagging

?

JuK

me2

I just wish it would use part of the file name to compare to the "suggestions" from the musicbrainz server to preselect the "most likely" one...

Berkz: Well, there is a value for relevance that's returned by the TRM server, but I think that's currently ignored.

being ignored by the client or server?

by JuK

But ideally that will be fixed in KDE 3.3.

cool

have u tried amarok?

yes

liked it?

Too flashy for me.

haven't tried it yet. Really like juk

are you running the beta2?

CVS

life on the edge :)

can you log into msn through kopete?

Don't know; don't care. ;-)

o.k

I haven't updated in a few days anyway though...

what distro?

a very old suse

But I meant KDE CVS.

Hmm... I've found TRM to be a little more unique than might otherwise be expected.

I know

i wonder

if I did a TRM on all the files I've ripped myself

honestly how many of them would turn up correct positive matches?

Pipian: Well, right now I'm getting about 50% drops from the server.

I'm currently running TRM through my collection.. 1/3 finds matches..

Well

I've calculated TRM's for many of my rips thus far

Berkz: It used to be a lot higher -- but right now the server is overloaded and about half of the time just returns with an error.

Pipian: what bugs me is that when I find matches I find 2-5 matches and I have to select which one :)

shame musicbrainz python doesn't work on Gentoo

blah

Berkz: Well, probably better than incorrectly tagging your music. :-)

wheels: true

ruaok: does that seem about on par -- 50% of lookups failing lately? Or is the code I'm working on just really b0rked? :-)

when using command : trm -m $filename.mp3 .. does the trm command contact the master server at all?

*goes to emerge gcj so libffi installs*

wheels: what kind of lookups?

ruaok: TRM

yup.

there server is overloaded. :-(

Nothing we can do about it right now.

ruaok: yeah, that's what I assumed -- but it's scarry when debugging code. :-)

I'm hoping to deploy a new server soon, but that won't help much.

ok, time to get off the computer for a bit. I need to relax a bit... OSCON was hard work.

Grar

when using command : trm -m $filename.mp3 .. does the trm command contact the master server at all?

I thought the TRM would only make a TRM signature and dump it on the console?

Berkz: yes

wheels: why does it need to contact the master server?

wheels: I thought it would only do that If I was to select the -lookup flag..

Berkz: I'm thinking that there's a wiki page that explains all of this...looking.

wikipedia.org?

Berkz: http://www.musicbrainz.org/docs/20031108-2.html

thx

ruaok: I seem to be getting some strange values back from eResult for the type -- it's a value greater than any specified in the enum...

ruaok: As such I'm having trouble resolving collisions.

MacIntire is now known as bkdelong

aha

I got a crappy ebuild of gcc-3.3.3-r6

that might explain why ctypes won't compile

Gotta love Gentoo. ;-)

apparently it didn't have the right use flags

so it couldn't possibly compile gcj

which is what ffi is compiled for.

thus ctypes couldn't work

that's one way to go about it =P

ruaok: Oh, another thing that I meant to point out -- you might want to run some of the example thingers and whatnot under valgrind -- I ran into a couple of memory errors there.

Woohoo. And now it seems that things are working...

ruaok: Oh, and I know you're not here at the moment -- but queing questions for batch processing later... Under which circumstances will tr_GetResults() resturn something other than eTrackList for the type?

Hi. Based on the supported calls, it seems that the exposed web-service API is meant for tagging files, not more general queries about the info in the db.

for example, there doesn't seem to be a good way to map artist->list of albums, even via a series of calls (unless one knows a track on an album to do FileInfoLookup

)

Which API specifically are you referring to?

the api exposed at mq_2_1.pl

i.e. the set of rdf calls responded to by the server (or really, any exposed way to query data; but scraping the data from the artist page doesn't count)

for example, in Artist.pm (used to display the artists albums via the web pages), there's a query into the database to compile the list of albums by an artist, but there is no way for me to query this information without downloading a dump (which I have done), or scraping the web page

uh, well I mean there's a pretty direct mapping between the calls supported in the mb_server project and those in the mb_client project to which you refer

in particular, the list of queries I mentioned in the Server queries section of include/musicbrainz/queries.h

all the MBQ_FOO calls

whether you're using mb_client or some ad hoc rdf code (the latter being my approach), there's still no way to get list of albums by an artist

because the rdf api seems geared only toward the mb_client client, which is meant for tagging-applications...it just seems like a richer api could be exposed...there seems to be nowhere else for me to discuss such an idea/proposal

i mean it's essentially trivial to implement in mb_server, and adding it to mb_client is just writing some menial hooks

(Cycles through all files cross-checking previously calculated but not-inserted-into-the-database-by-me TRMs against results in the database to figure out whether or not the algorithm results in more or less unique hits (Of course it may be offset by luck of people having the same encoding as me)

Will also be helpful in identifying TRMs to *add* =P