Hi there!
I’m a newbie to Hooktheory, but it looks amazing!
I’m working on a key detection comparison test, to see how well certain apps perform accuracy-wise.
In order to determine, I think Hooktheory is a great resource, to gather all the youtube snippets from the TheoryTab database.
Then let these apps determine them key-wise, and cross-reference if they got it right or wrong (assuming Hooktheory is 100% right).
Few questions come to mind:
-
The database as of right now, how is the key-determined?
Is that based on the notes filled in by the user?
So for example someone fills in C – G – Am – F as melody, Hooktheory itself determines that’s a song in C Major? Or is this something the user fills in? -
I’d love to have a free-from-pollution database ofcourse, however, I do come across some questionable examples:
yeeee by yeah aightt Chords, Melody, and Music Theory Analysis - Hooktheory : the melody and the youtube video don’t match
Sergei Rachmaninoff by 12 Romances - No 7 (Arr Volodos) Chords, Melody, and Music Theory Analysis - Hooktheory - same here
Can you guys determine what’s going wrong?
Maybe trolls (though unlikely, since you pay for your service) could simply fill in some randomness?
Any rough estimate how much of such ‘off examples’ are in the total db?
Then I can simply extract a certain margin of error in my test.
Thanks!
1 post - 1 participant