Automatic Key Detection in Audio Recordings Using the Signature of Fifths

The article presents new approaches to determining the key of a musical piece in audio format. These approaches build upon key detection algorithms that utilize the signature of fifths, which have previously been applied to musical pieces in symbolic form (MIDI). To adapt these algorithms for audio signal processing, short-term power spectral analysis was used to determine the signature of fifths corresponding to a given piece. The experiments were conducted on a set of 47 piano pieces by composers from various eras, recorded on a hybrid acoustic piano equipped with an add-on that enabled the production of synchronous symbolic notation. This setup made it possible to compare the accuracy of the proposed audio key detection algorithms with that of their symbolic precursors. Additionally, the comparative analysis was supplemented with experimental results obtained using a strictly correlational key-detection method implementing various tonal profiles.