Logo image
Back
Can I Have the Keys?: Key Validation Using a MIDI Database
Conference proceeding   Open access

Can I Have the Keys?: Key Validation Using a MIDI Database

Joshua Albrecht and Daniel Shanahan
Proceedings of the ICMPC-SMPC 2016 Joint Conference, pp.752-755
2016
pdf
Albrecht_Shanahan_Keyfinding_validation631.90 kBDownloadView
Open Access Free to read and download
url
https://www.scribd.com/document/462861607/ICMPC14-Proceedings-pdf#page=799View
Full conference

Abstract

The accuracy rates of a recently-developed key finding algorithm were estimated to be quite high, higher than most other currently available algorithms. This study provides a) an external validation of the accuracy rates of that algorithm on an independent dataset, and b) a validation of that dataset using the algorithm. Specifically, a large dataset of over 12,000 MIDI works found on ClassicalArchives.com were sampled for this purpose. MIDI data tends to contain a high percentage of errors. To determine whether the database provides a sufficient signal to noise ratio for pitch-reliant tasks like key-finding, an initial validation study was performed on the dataset in which accuracy rates for the key-finding algorithm on a small subset of 32 works were compared with a ground truth of accuracy rates. In the initial study, it was found that 28 of 32 (87.5%) of MIDI works returned the same result as the ground truth. Given this relatively high accuracy rate, it was determined that the database can be an appropriate vehicle for testing a key finding algorithm. On a dataset of 192 separate works, it was found that the algorithm correctly assigned keys to 164 works (85.4%). Together, the two studies are consistent with the ideas that, 1) though there will be some errors as a result of noisy data, a MIDI database can serve some uses for some studies, and 2) the key finding meta-algorithm provides an accurate way to assign keys to this database.

Details

Metrics

1 File views/ downloads
1 Record Views
Logo image