2 Comments
User's avatar
⭠ Return to thread
Chris Dalla Riva's avatar

Mostly just because the dataset didn’t note the key. Because of that there was no reliable way to assess relative positions. Had that been part of the data, I probably would have done that

Expand full comment
Skoddy's avatar
7dEdited

Yes ben's comment is what I was interested in as well. As a musician and songwriter, the relationships between the chords (progressions) are more important than the actual letter names or use frequency.

There's likely an algorithm that could be applied to the dataset to determine the key with some accuracy. I always look for the 2 major chords in a song that are consecutive to each other (they don't need to be played consecutively), those are usually the IV and V chord which leaves the other major chord as the tonic, e.g. C and D are the IV and V leaving G as the key. Of course this isn't a rule but does apply to the majority of popular songs. I'm sure some software engineers out there have more specific algorithms.

Or cross reference with another dataset like https://tunebat.com/

Expand full comment