How To Place Songs On Appropriate Scoreboards?

[blue_h3x]

I think S was suggesting numbers at the start of song titles should be dropped. Although why wouldn't this apply to 99 Luftballons? The number would be dropped by the suggested rule which would result in a wrong title.

[MarimbaSurf]

I think S was suggesting numbers at the start of song titles should be dropped. Although why wouldn't this apply to 99 Luftballons? The number would be dropped by the suggested rule which would result in a wrong title.

And 42 by Coldplay would crash the matrix.

[S.]

I think S was suggesting numbers at the start of song titles should be dropped. Although why wouldn't this apply to 99 Luftballons? The number would be dropped by the suggested rule which would result in a wrong title.

99 is part of the song title. I was refering to the track number that sometimes gets put into the song title.

[MarimbaSurf]

99 is part of the song title. I was refering to the track number that sometimes gets put into the song title.

You'll still have to find a method that is able to decide whether a number is supposed to be a track number or a part of the title.

[Uupis]

How about S' method supplemented with user verification?
i.e. before the track starts, like during character selection or before clicking "Play" you can see the title and some sign that the title is corrected. If the title is wrong, you selet the correct one.

[S.]

99 is part of the song title. I was refering to the track number that sometimes gets put into the song title.

You'll still have to find a method that is able to decide whether a number is supposed to be a track number or a part of the title.

Which is what I brought up right away in the earlier post, but people seem to have misunderstood.

[MarimbaSurf]

Can't really find it, oh well :|

In case it implies looking for dashes, be careful with track names like "9-bit Expedition".

[blue_h3x]

One way would be to look for scores submitted without the number, although that'll require some kind of trend spotting or learning algorithm which is way ott for this.

As suggested, a simple 'Did you mean ... ?' would suffice in more cases.

[Dylan]

I think it will work well enough to just drop all leading numbers and dashes (and dots).   That would still likely be a unique title for the artist, especially with the other stats like density and duration being tested for.

My current thinking is this:
--convert to lowercase
--convert " and " to "+";
--convert "&" to "+";
--remove all whitespace
--remove umlauts (for example, "ö" becomes "o");
--remove digits from the beginning until a non-digit character is reached
--convert "featuring" to "feat."
--convert "remix" to "mix"
--remove these: ".",",",">","<","'","´","`","(",")","[","]","-","album","version","the","original","amazonexclusive","soundtrack","explicit","disc","cd","single","bonus","track","edition","deluxe","main","musicalscore"
--remove all digits

--It saves the string between some of the steps so if one of them (like removing all digits) leaves too little to work with the result of an earlier step is used instead.

--These steps are applied to both the title and artist.
--removes the artist name from the title if it is in there.
--last, it combines the title and artist into a search string for the track

You can test it here:
http://air.audio-surf.com/as/airgame_findaddsongid.php?song=02 - Keyboard Milk&artist=Röyksopp

The modified title and artist are only used by the scoreboard server and players never see them.

[MarimbaSurf]

These methods are going really far, so that a collision becomes very likely.

Edit: Found some. 'Open Goldberg Variations'. Here are some examples:
Variatio 13 a 2 Clav.
Variatio 14 a 2 Clav.
Variatio 17 a 2 Clav.
and so on will all have the string 'variatioaclav'. I'm going to check my entire library tomorrow.

Edit2: Also 'Robot Waiting Room #1' to 'Robot Waiting Room #5' from Portal 2.

[blue_h3x]

These methods are going really far, so that a collision becomes very likely.

Can you find some?

Although this method would make live mixes appear with album mixes would it not?

[Mincus]

How will this affect unicode tags? (Yes, I know we're the bane of your existence )
For artist 浜崎あゆみ (that is, Hamasaki Ayumi) and song BLUE BIRD I currently get bluebird浜åŽã‚ゆみ spewed out. Will that affect what is shown to the player on the high score screen or is it purely for internal checking? Edit: (I missed the "scoreboard server" line)
Will that affect anything internally? On the surface it looks not to.

Messing with things to that extent may cause collisions within the same artist, especially with game music tracks that may not have names and may be purely numbered as "Track 1", "Track 2" etc.

It's possible pulling in the song length will help with this to some extent.
Or I had an idea with the rollercoaster that may work: The rollercoasters are usually fairly similar for the same track and the displayed ones for the scores have a much reduced data set than the entire track, so it may be possible to compare those data points with a margin for error (in all 4 directions) to establish whether the tracks are the same.
This does add another problem though: Do you just abandon these tracks with no highscore as "I can't find one" or do you do more extensive checking for similar names and do the rollercoaster test against those.
Obviously this is a lot of work, but I thought it's an interesting idea and if the lower-quality analysis is implemented, that should bring the rollercoasters closer together.

Other than those two suggestions I don't know if there's a smart way around this without doing something like the fingerprinting last.fm has or outright asking the user with a list of suggestions. Rhythm Zone uses last.fm (or used to before the servers stopped working) so it's definitely possible to squeeze it into Unity somewhere and might be worth looking at.

Or we might just have to accept that the system isn't going to be perfect and take the hits on things like game music tracks (which I will, but I'm not a competitive player).

Edit2: Make sure you can undo the "remove the artist name from the title" as well, some tracks purely have their artist as the title, example: Iron Maiden - Iron Maiden.

Edit3: David Bowie's two tracks Fame and the remake Fame '90 collide under the current rules. I don't think this is a rare enough event to write off, unfortunately as I'm sure there are many similarly named remakes of tracks by a multitude of artists.

[blue_h3x]

Other than those two suggestions I don't know if there's a smart way around this without doing something like the fingerprinting last.fm has or outright asking the user with a list of suggestions. Rhythm Zone uses last.fm (or used to before the servers stopped working) so it's definitely possible to squeeze it into Unity somewhere and might be worth looking at.

http://www.freedb.org/ offers a similar service, the game could fingerprint the tracks as Mincus suggested using either service and then use the supplied track title and artist for scoreboard look up. In the case of more than one being returned from the service, you could compare to the one the user supplied ie ID3 tags etc, or simply ask.


As for song collision, from the album 'Systematic Chaos' by 'Dream Theater' the first track is called 'In The Presence Of Enemies - Part 1' and the last track is called 'In The Presence Of Enemies - Part 2'

Track 1, In The Presence Of Enemies - Part 1, is 9 minutes long.
Track 8, In The Presence Of Enemies - Part 2, is 16:38 long.

Obviously these two will produce very different rides and score, yet the look up will nuke them to the same scoreboard.


EDIT:

Edit2: Make sure you can undo the "remove the artist name from the title" as well, some tracks purely have their artist as the title, example: Iron Maiden - Iron Maiden.

Also 'Avantasia' by 'Avantasia' will collapse as well.

The Album 'Funeral' by 'Arcade Fire' has four tracks that may suffer a collapse, named 'Neighborhood #1', 'Neighborhood #2', 'Neighborhood #3' and 'Neighborhood #4'.

[S.]

Yep, basically any song that consists of multiple unnamed parts will collide.

Many bands have a song named after themselves. Then there's songs that are made up entirely of numbers (cant name one right now, but i know they exist), but I think you mentioned you already fixed these two.

What are you gonna do about songs with [live] tags?

On another area, what if the ID3 tags are not filled and you have only the filename to work with? (UT2004's soundtrack does this, they make interesting surfs)

[Passerby]

Elliott smith, has "waltz no.1" and "waltz no.2", and there is "10.000 Days", "4°", and the 2 songs named "useful idiot" and the song "Message to Harry Manback appears on 2 albums, slightly different on each, but tool.