beaTunes News

Monday, September 17, 2007

Explaining Rulesets

The other day a customer sent us an email with a couple of questions that are asked again and again. I would like to take a few minutes and try to clarify a number of things.

What determines the color?

beaTunes looks at a part of each track and performs a frequency analysis of that part. The result is then projected into the RGB (color) space. In essence, it is a measure for whether different songs are using the same acoustic frequencies. The idea is that songs that share similar frequencies are more likely to sound alike. So two acoustic guitar singer/songwriter tracks are more likely to have a similar color than an acoustic guitar song and some AC/DC anthem.

But same color definitely doesn't mean that two songs sound the same. They only share some acoustical properties. It is meant as one of multiple factors that can be used to find matching songs. It is by no means the all determining factor.

Which leads us directly to the next question:

How do I configure rulesets?

Let's look at an example. Say you have three rules, a), b) and c):

  • a) set to 1
  • b) set to 4
  • c) set to 2

If one song matches another song for all three rules 100%, then this song will be awarded 1+4+2=7 points. So in this case 7 points equal 100%.

If a song only matches rule a) and b), that would be 1+4+0=5, and that would be 71.4% (5/7).

So in essence you weigh the rules. In our example, rule a) has very little weight (1) - so it's really not that important. Rule b) on the other hand determines most of the outcome. It alone can score more than 50% (i.e. 4/7).

Note that matching rules isn't necessarily binary. That is, a song can match a song partially. So if a song matches rule b) somewhat, it might be awarded 2 out of 4 points. A good example for this is key matching.

How does key matching work?

Let's assume the key rule's weight has been set to 1. If it's a 100% match, i.e. a song has the same key as the one you would like to match, then beaTunes awards 1 point. If a song is in the dominant or subdominant of the song you would like to match, beaTunes awards 0.75 points. If it's in the tonic parallel, beaTunes awards 0.5 points. If it's neither of these three, beaTunes awards 0 points.

Please note that for key matching to make any sense you first have to set the key for a significant number of songs. beaTunes is capable of reading existing key values from id3 tags, but is not yet able to automatically determine the key.

What about negative values for rules?

Again, say you have three rules:

  • a) set to 1
  • b) set to -4
  • c) set to 2

In this case the total number of points a song can score is 1+0+2=3 (the negative weight is ignored for this).

The values for the matching rules are added up. So, if a song matches all three rules the score it reaches is 1-4+2=-1.

In essence, with this setting you want to make sure that a song that matches rule b) is ranked really low.

What kind of similarity does the rule “similar tags” mean?

beaTunes lets you tag tracks with keywords (both in the Get Info dialog and in the table view tags column).

Again the -5 to 5 slider determines how important this rule is relative to the other rules. Assuming it's set to 1, if two songs have the exact same tags, 1 point is awarded. If two songs have two tags and one of the two tags is the same, 0.5 points are awarded. etc.

What is the difference between the “match quality” slider in the preferences and the one in the matchlist dialog?

The slider in the preferences applies only to the the "matching songs" panel (Edit > Show Matching Songs). The one shown when creating matchlists applies just to the matchlist you are creating. Check out our demo video for creating playlists using the matching songs panel.

I hope this post explained one of the core features of beaTunes a little better and helps you build some great playlists.

Labels: , ,


Post a Comment

<< Home