In the Billboard dataset, we have finished the process of translating our previous **harmony chord symbols into roman numerals.
Specifically, we are using a slightly modified version of humdrum's standard **harm interpretation.
In order to create accurate roman numerals, we made numerous small fixes to the harmonic transcriptions, mostly regarding the keys of the transcriptions. The original McGill harmonic transcriptions often omit key changes, an in a few cases simply mislabel the key. They also sometimes mispell chords given the key---for example, spelling a chord Db minor in the key of E major---which leads to weird roman numerals. There are still some places where interpreting the key or how to "spell" a chord in the key is genuinely difficult (check out "Do I Do" by Stevie Wonder for example!), but for the most part we've fixed things so the roman numerals come out reasonable.
We have also taken some steps to keep the "original" **harte annotations in an exact one-to-one relationship with symbols in the McGill "Mirex files."
Most notably, we have pulled all of McGill's timestamps into our files, so nearly every chord has a timestamp associated with it.
We've incorporated updated (better aligned) versions of the Billboard valence-arousal subset files.
The dataset is now almost entirely complete and consistent with what we described in our 2023 paper in the proceedings of International Conference for Music Information Retrieval (ISMIR 2023). We have a few things to wrap up, but should have those finished in the next month.