SongKong Jaikoz

SongKong and Jaikoz Music Tagger Community Forum

Ignore Musicbrainz Compilations - Prefer Original Release - Does it Work?

The check box to prefer original release even if the meta data match is better for a newer release doesn’t seem to be very effective.

I’ve experimented with this setting on and even check the boxes so it would only match on Acoustic ID, and/or set the Meta data match minimum very low to 30 or 40. Most of the time it doesn’t get the original release even though you can see it if you do a manual Musicbrainz match.

So I did manual matches on several greatest hits tracks from several artists. In looking at the list, if Jaikoz just ignored compilation matches and matched to the first/best matching album, it would work like a charm in many cases. But it appears the prefer original release checkbox isn’t doing this.

I’d love it if this option could be added. I want the album and date on all my tracks (except remixes) to be the original album and date. Looks like the only way to do it is manually so far. That’s a lot of work with over 1,000 tracks on compilations.

Also I assume Jaikoz doesn’t overwrite the track length in the tag. A lot of times the compilation version is a different version (shorter/longer) than the original track. So effectively I want the metadata from the original release and the track length to be correct. I think Jaikoz can do this. Just want an option to do it automatically (ignore compilation matches)

Hi

The beta release of Jaikoz is going to be way more effective at not matching compilations if you so desire, and will only match to them as a last resort. I asked whether there was a requirement to Never allow compilations http://www.jaikoz.net/jaikozforum/posts/list/1523.page

and the only answer back was No, but would you like that ?

I also posted a question http://jaikoz.blogspot.com/2010/09/matching-to-different-song-version.html about songs lengths, and whether anybody would want to match to a version of a song that was an earlier release even when it clearly was a different version because of large discrepancies in the track length, and the only answer was also No, but would you like that as well ?

That would be awesome. I’ve been playing with the matching a lot. One example is Madonna, Frozen, from Greatest Hits. Jaikoz doesn’t even match the compilation and all the existing tags for the song are correct. The best match is 39 but to the song Ray of Light.

However if I go to the Musicbrainz website and do this advanced track search:

track:frozen artist:madonna type:album

It matches with score 100 to the correct original track on the Ray of Light album.

I thought a lot about this. Instead of an “Ignore Compilation Matches” option I think this option would be much better:

“Match to earliest non-compilation release ONLY (album match is ignored)”

I’d make the match score based on artist, track title, lyricist and conductor only (and maybe the release country) and give artist and track title 3 times the match score weight as lyricist, conductor, and country. Then choose the match with the earliest year that meets the user’s match minimum. I’d also ignore the AmplifindID match settings as most likely it won’t match if it is a different release in many cases and the title match in the metadata is really the important thing anyway.

If this is too much overlap with what you already have, then yes, a separate ignore compilations would be desired. But I think the above is a better way to do it. Basically, you just don’t want to figure the acoustic match, album title, year, and track no. in the match when trying to match the original release.

There may be some debate though on using acoustic match. Some compilation releases are very different or remixes of the original - but I think that is overcome by matching the title. Frozen (Techno Remix) should not match to Frozen the original release. Maybe you need to throw the remixer field into the match? But is that even populate for most remixes?

Anyway, how useful this will be as then I can easily see what year and album the song was originally on instead of looking at Best Of, Greatest Hits, bla bla bla all the time.

My opinion is:
if there are two songs - one original release and one compilation - so the same artist and the same title, but with a different track lenghts of more than plus/minus 10 sec, then these songs are different:!:
What I will say, a compilation song release, which is longer or shorter than 10 sec than the original release, should never match to the original release.
If someone want a match to the original release, should buy the original release :!:

That makes sense. But I don’t see the harm in offering the option for those that want to be able to match to the original release no matter the track length. Option “Match original releases with exact same title and artist but significantly different track length”.

This may be an option that the majority of users don’t need but I’ve found the best software, such as MediaMonkey is so successful just because of the fact they offer features/options that appeal to the widest audience take everyone’s needs into account.

I was going to suggest that there is either a length field in the table that is highlighted yellow in the case the match track has length more than +/- 10 seconds so the user can decide whether to match it or not, or maybe a pop-up for them to go through and confirm each track that the length is significantly different.

Is there a way to revert to saved version on a per track basis? If not, that would be useful.

Also there seems to be a bug in that if you uncheck the box to not lookup online tracks that already have an MB id, when you go back into preferences, it is checked again. This forces you to have to manually delete the MB id from the track if you want to look it up again.

I may offer the option.

As for how you do it, the algorithm for matching has been improved significantly and should work well in the majority of cases. The AUtocorrection is designed so that no user intervention is reuired.

To revert a track, right click on a field in the track and select Save/Revert to Saved

I’ll check the Mb id Preference, but really you wouldn’t want this set to unchecked by default, I was actually considering removing this option.

My initial test seems to show that Jaikoz still won’t match to original release when it should. Test example:

Options set:

Do no match online if already have a MusicBrainz ID (not checked)
Do not update from Discogs when matching (not checked)
Retrieve acoustic IDs as required (not checked)
Prefer do not match to various artists (checked)
Prefer do not match to single artist compilations (checked)
Country of Release: USA
Preferred Release Date: Earliest Release Date

I cleared the acoustic ID and musicbrainz ID and did an Auto Correct. Here is results:

“Dancing Queen” - ABBA - The Best of ABBA 20th Century Masters - 2000
Track is 3:49 long original release on album “Arrival” is 3:50
Matches to SOS on the same compilation album? Huh?

“Mamma Mia” on same album matches to “Waterloo” on same album

But…

“Already Gone” - Eagles - The Complete Greatest Hits
length:4:15

Matched to correct original release from album “On The Border” length:4:16 hurray

But when I did a match for the entire greatest album, it did like abba and started matching to completely different songs. It would seem to me the title should carry some very very heavy weight for the match. I don’t see when there is no MBid and no Amplifind ID, how in the world it would match the song to a completely different title???

If these songs do not already have puids you should check Retrieve acoustic IDs as required , this allows Jaikoz to match the tracks to songs from different album s thaen your medtadata indicates.

Also I dont understand why you cleared the aciutsic id, you should only have cleared the musicbrainz id.

I erased the acoustic ID as I wasn’t sure how precise it is but sounds like not that precise so a slightly different length track will still match I assume?

It still doesn’t explain if I have a correctly tagged track that is within a second or two of the original release and no acoustic ID, why would it match to a totally different song on the same compilation album?

I tried a manual MB match and on the Dancing Queen test it only shows Dancing Queen title. But if I use Auto match it matches to SOS?

I retrieved the acoustic ID for Dancing Queen and it still matches to SOS.

At one point I thought, maybe it IS SOS? Nope, it’s Dancing Queen.

I actually saw this bug on the current production releases. It happened when I used manual match, selected the song to match and then it matched it to a different song by the same artist.

Sorry I’m not following your what saying here.

Firstly retry try the Automatic Match ensuring you have enabled retrieving acosustic ids , and reinstate the acoustic id. If it doesn’t give the results you want then send me your support files using Advanced/Create Support Files this will show me the decisions taken by Jaikoz to reach its finsl decision.

Secondly, can you restate the problem with the Mbids of what you wanted the song to match and what it did match.

Thirdly, not the Manual Match uses the old algorithm and is due for a complete rewrite, so dont expect the manual match results to concur with the Automatic match.

Automatic matching on the beta is done very differently to the production release so although you might be seeing the same problem the cause will probably not be the same.

Sorry I missed the option question threads regarding compilations because they were asked before I came across Jaikoz. Had I been a member then I for one would have said a resounding YES - we do need the option!

What I wanted to say has already been said earlier:

“This may be an option that the majority of users don’t need but I’ve found the best software, such as MediaMonkey is so successful just because of the fact they offer features/options that appeal to the widest audience take everyone’s needs into account.”

Personally, Jaikoz doesn’t do exactly what I want yet, but it’s way, way closer than any of the other taggers out there.

The point is that everyone stores their MP3 music collections in different ways. Some of you may just collect albums - in which case any tagger will do a reasonable job of on-line look-up.

I don’t. I’m trying to sort out the MP3 library for a radio station - basically a terrabyte hard disk of single files - and a few albums.

Most are in folders like ‘Sixties’ (which contain all the chart single releases between 1960 and 1969) and ‘Seventies’ (releases between 1970 and 1979)… and so on. There are 30-40 or so other folders like Rock, Motown, Country, Reggae and so on - each containing up to 10,000 MP3 tracks.

The folder I’m working on at this moment in time is called ‘Unsorted’ and that alone contains 30,667 MP3s!

My first priority is to move all the 60s, 70s, 80s etc. tracks into their respective decades folders and to do that they need to be dated with the correct year of release. There must be many duplicates and incorrectly tagged/named files in the wrong locations.

So to me, Jackie Wilson’s Reet Petite is a 1957 song. I don’t give a toss that it was re-released and got to number 1 in 1986 - or that it has appeared on countless 1950s compilation albums in the years since it’s original release.

It’s also of no interest to me that I just happen to have a version that came from one of these later albums and is a different length - because I want the song to be looked up using it’s filename only, ignoring the length, size and acoustic fingerprint of the file.

If a listener phones in and wants to hear that Jackie Wilson song, they (or we) aren’t going to be all that bothered which particular version it is that gets played.

Likewise, if our playout system is asked to generate a random playlist for a show with music from a particular decade, we want Reet Petite to only get played in the 50s show.

The new beta version of Jaikoz does indeed go a good way towards making what I personally need in a tagger possible - without a lot of tedious tweaking, and from my personal requirements, the only option I can see missing is quite simple:

[] Always Select Oldest Year On Remote Look-Up

With that option enabled, any incorrect years for the original release dates will be down to the online database - not Jaikoz.

With the current limitations of most taggers in mind, I’ve actually written my own MP3 dating software which I can use after using Jaikoz to correct faulty release date years returned by the on-line databases.

It uses it’s own built-in database, but it’s still quite small at the moment (it currently only has 22,455 songs in it).

It works perfectly, but at the moment only does exact matches using the MP3 filename format ‘Artiste Name - Song Title’ (non case-sensitive). A missing apostrophe in the word Don’t will currently not find a match.

To avoid this, I’m now writing a ‘fuzzy’ match function which will match songs even with minor differences.

Let me know if anyone else would find this program useful for use in conjunction with Jaikoz. If enough people are, (with with Paul’s permission), maybe I could start a thread for it elsewhere.

TenBaz

I’m interested

Please see my summary here

http://www.jthink.net/jaikozforum/posts/list/1588.page#6770

[quote=TenBaz]
It’s also of no interest to me that I just happen to have a version that came from one of these later albums and is a different length - because I want the song to be looked up using it’s filename only, ignoring the length, size and acoustic fingerprint of the file.

If a listener phones in and wants to hear that Jackie Wilson song, they (or we) aren’t going to be all that bothered which particular version it is that gets played.

Likewise, if our playout system is asked to generate a random playlist for a show with music from a particular decade, we want Reet Petite to only get played in the 50s show.

The new beta version of Jaikoz does indeed go a good way towards making what I personally need in a tagger possible - without a lot of tedious tweaking, and from my personal requirements, the only option I can see missing is quite simple:

[] Always Select Oldest Year On Remote Look-Up

With that option enabled, any incorrect years for the original release dates will be down to the online database - not Jaikoz.

With the current limitations of most taggers in mind, I’ve actually written my own MP3 dating software which I can use after using Jaikoz to correct faulty release date years returned by the on-line databases.

It uses it’s own built-in database, but it’s still quite small at the moment (it currently only has 22,455 songs in it).

It works perfectly, but at the moment only does exact matches using the MP3 filename format ‘Artiste Name - Song Title’ (non case-sensitive). A missing apostrophe in the word Don’t will currently not find a match.

To avoid this, I’m now writing a ‘fuzzy’ match function which will match songs even with minor differences.

Let me know if anyone else would find this program useful for use in conjunction with Jaikoz. If enough people are, (with with Paul’s permission), maybe I could start a thread for it elsewhere.

TenBaz[/quote]
Hi TenBaz whilst admirable I think you would be better use of your time adding more information to the Musicbrainz database rather than your own, you may find that much of the information is already in there and you would just have to fill in the missing nits.

If you did this then Jaikoz would work for you except for the occasions where your song in indeed a different version, but this functionality will be added in the fiirst post NGS release of Jaikoz see my summary here.

http://www.jthink.net/jaikozforum/posts/list/1588.page#6770

Actually, I’ve got to the stage where I’ve done too much work on my MP3 dating software to stop now - there are currently over 26,300 tracks in the database.

I also believe that finishing my program would probably be a lot easier than doing the same amount of work on any (or all) of the on-line databases.

Users can use Jaikoz to sort everything out and then run my program afterwards to correct the year tags (no other ID3 tags are altered).

I would however, in the future, like to give users the ability to add to the database as it’s obviously a lot of work for one person, though I’ve decided to stop short of making the finished database available online for software like Jaikoz to query as I can’t find any info on how to do it.

TenBaz

I think the exact match on Artist Name and Song Name would cause problems for other users because actually in many cases it is unclear what the song name or artist name should be, for example take a look a look at the links on http://wiki.musicbrainz.org/Main_Title. That is why Jaikoz does a fuzzy search match rather than exact search match.

I can appreciate why you wouldn’t want to put all this information into Musicbrainz yourself but it would be great if you did make the database available so perhaps others could. If you like I could help you with this , even hosting it on my website if you like, what format is the ‘database’ in ?

Absolutely. That’s why such an option in Jaikoz would only be of use to those of us who know that our tracks are correctly named. There must be a few of us out here that do, though I agree that the vast majority won’t.

That’s why I thought that the best option would be to use Jaikoz to sort out all the MP3 filenames and ID3 tags and then use another program (mine) to run through them all again fixing the release date years.

Simples! :slight_smile:

My program does too. The database is optimized for matching by having everything but spaces, a…z and numeric characters stripped out of it. The word ‘the’ is removed from the start of artiste names and all ‘&’ characters are replaced with the word ‘AND’.

On lookup using ‘exact match mode’, the program applies the same rules to the MP3 filename - and invariably a match is found.

In ‘weighted’ mode, an algorythm gives the filename a weight value and displays all entries in the database with a similar weight value, though my algorythm needs a little more work as it is currently a bit more hit and miss than I would like! :slight_smile:

I’d need to make the ‘database’ available to others - primarily because I want others to add to it (I’ve spent days adding the 26K+ in it at the moment) - but I have no idea what the best way to go about it is. I’ve asked for advice on a number of other forums and so far had no replies.

So, at the moment until I get this info, I’ve left it all in a plain ASCII text file which uses the basic format:

Artiste Name - Song Title - Release Year

(To speed up my program, I may add the calculated weight value to the line so it doesn’t have to be calculated each time you do a search).

If I was to add more ID3 tag fields then my program starts turning into a full tagging program and we already have the best one out there! :wink:

In a nutshell, I don’t think hosting the DB is the problem - it’s more a case of not knowing what format database it should be and what else is involved - for example, with regards to handling queries from clients.

I want others to add missing tracks to the database, but I assume I need admins to check the submitted data before it is accepted - like Wiki I presume…

Otherwise it’s a case of having the user submit new entries to me via email from the program and I add them to the database myself.

If it doesn’t need to be online then I could convert it to anything I suppose (Paradox is preferred to integrate the easiest into a Delphi app), but doing that makes it more difficult for others to add to the database.

So, any suggestions would be most welcome.

What I will do in the meantime is let you have a copy of the first test release version of my program (including the ASCII database as it is at the moment) so you can take a look yourself.

TenBaz