Extracting the key from music

Abstract

Extracting a sense of key from music audio is indispensable for various unequalled end-user applications dealing with music playback. This chapter presents an audio key extraction algorithm that is based on models of human auditory perception and music cognition. It is straightforward and has minimal computing requirements. First, it computes a chroma spectrum from short non-overlapping time frames of audio; a chroma spectrum represents the spectral energies collected over all pitches that share the same chroma. This chroma spectrum is compared with profiles for all existing 24 Western keys; a key profile represents the perceived stability of each chroma within a given key. The key profile that compares best with the computed chroma spectrum is taken as the most likely key. An evaluation with 237 CD recordings of Classical piano sonatas indicated a classification accuracy of 86%. By considering keys that are 'friendly' to each other as equal keys, the accuracy is even 96%. © 2006 Springer.