#musicbrainz

we need more vocal ARs. I'm up to 16 other vocal ARs on this album

not sure if this is the place to ask about picard but, when i lookup a few of my albums it does not seem to correct the tags on a few songs which i'm assume thats because my tags were right on those ones. The issue is that even though theres no correct to make on those tags i still want picard to rename the files based on the existing tags like it does for the others. Anyway to force this behavior?

luks you there?

luks: I am at the point of building picard-qt now. I run sudo python setup.py build_ext -i

yep, i'm here

why sudo?

one min

and get an error

ofal/ofa.h

not found

you can safely ignore that

ok

will keep going

the fingerprinting part is not fully implemented anyway

ahh

you can ignore the gstreamer one too

you would need to install libofa and gstreamer

i see..

but i can ignore

gregf: right click on the album and choose force savve to see if that changes if from a check-mark to a bar.

*save

ok have it running

I will now check it out

:) Thanks for the help

but be careful

it's really not a finished app yet

kerensky97: tried that, had the same idea earlier

I will only use it on copies of my music :)

kerensky97: seems to stay blank, no metadata under the id3 on the bottom tab area, butunder the musicbrianz tab all the data seems to be there if it helps

luks: do you use KDE or GNOME? or Other?

GNOME

and Windows XP :)

theres a smile after that ;)

i usually say it with a :( after

well, i'm happy enough with windows

but i have a new machine and no free win license

yeah

so i installed linux/gnome on it

where is your eclipse installed?

kerensky97: yeah when i do force save nothing happens i just double checked

uhh, dunno

where did the ubuntu package put it

no i mean what machine

ah, on both

win or ubuntu

luks: How is the GUI design work done with elcipse and pydev.. no glade or anything like that?

qt gui is usually done in the qt designer

gregf: I'm not sure if it's still not allowing you to save, you might want to just wipe the tags completely on the tracks having problems are then add tags from MB.

luks: so the *.ui files are the GUI as output from the designer?

yep

then they are compiled to a python code

kerensky97: yeah it def tagged the files just not renaming some i'll have to report it as a bug and see what they tell me

kerensky97: thanks though

luks: how is the GUI "hooked" to the python modules? For instance when i click the "add files to tagger" button, which function in which module is executed?

JaXxon_: it's based on qt's signal/slot architecture

in the code you connect signal clicked() from that button to a function

I am somewhat farmiliar with signals and slots

ok

gregf: you mean the files are correctly matched, but picard doesn't rename them?

luks: what plays a bigger role in identifying a track, the known metadata or the content id from musicdns?

currently the two are completely separated

you can do either metadata loopups or fingerprint lookups

when i choose analyse in picard that is doing a fingerprint?

yes

and if the puid matches more than one track, it compares also the metadata

Have you ever used foosic?

libFooId

If no puid match occurs, how is a new puid created? Is this only done by musicdns?

yes, new puids can be created only by the musicip mixer

and no, i haven't used foosic

they call it completely free, but the server isn't more then than musicdns...

err, *more free

So the "encode" is "free" but the search algorithim is not ? on both

yes

I guess the idea is more that foosic's algorithm is public?

Is musicbrainz one of the bigger tagger site?

And freely usable.

Yea i have looked at the foosic algorithm

srotta: the fingerprinting code is free in both solutions

But no where is any discussion about how you would search a collection of bits that represent finger prints

the only difference is that foosic is under BSD, libofa is GPL

luks: Oh, I thought the MusicIP algorithm was closed, sorry.

Does anyone know how the music dna (pandora) project works?

they manually "tag" each song

wow that's a lot of work

yeah

but it's also pretty accurate

yea it is

far more accurate then fingerprinting or last.fm-like playlists based on users stats

*than

Has there been much discussion about how the fingerprint search occurs?

depends how you define "accurate".

if i pick one song, i want a playlist that will sound similar

not based on bpm, frequency spectrum or statistics, but on the actual music

FWIW, foosic fingerprints have a well-defined difference function

like levenshtein distance

so searching them is viable using dimensionality-reduction and other such trickery

I've tried Pandora a bit, but I'd have to listen too much terrible crap to make it learn I haven't bothered :P

so yes, it is a lot more open than PUID or TRM, in that the fingerprint actually represents the music, and isn't just a reference to the actual location in the server.

russ_: i was talking about ofa fingerprints, not puids

to use foosic in MB like environment, you would need a server and short IDs anyway

true

but the MusicIP central server is not open-source either

but then the client could create the foosics

yes, it isn't open source. but at least there is one, and it works

foosic is a one-man project, afaik

moin kids!

moin

luks: :-)

russ_: did you ever setup the mayhemchaos account?

or just the mayhem account? :-)

WRT Pandora: we think that describing music through listening similarities is a lot better in the long term than describing music by an arbitrary number of arbitrary characteristics (Ambiguous vocals, wtf)

mayhemchaos is a subscriber, yes

russ_: the problem is that people listen to crap :)

russ_ is now known as Russss

the tag radios would be cool, but again people don't use them correctly

they can be improved by tweaking our algorithms, we aren't anywhere near doing that well.

it isn't that rare to hear metallica when you listen to a "hip-hop" tag station

and there is too few genre specific tags

we have an algorithm which tries to work out if users tagged stuff maliciously, unfortunately it still sucks

Is there nothing that we can tell about the "style" of music from its analog wave form?

yes, lots

Is anyone working on anything like that Russss

yes, we have a whole one-man department working on fun things like that.

:)

Are you that man?

god no

who is?

we have a music research guy

at mb?

at last.fm

there's some light reading

it was the winning paper in the 2005 MIREX genre classification competition

Russss: you are the last.fm guy right?

yes.

check the hostmask ;)

although someone took away my lovely reverse-DNS and gave me a freenode cloak :P