Influences of Signal Processing, Tone Profiles,and Chord Progressions on a Model for Estimating the Musical Key from Audio

摘要

We have investigated the effects of low-level DSP parameters, tone-profile values, and observation type for an HMM-bascd key-estimation algorithm on two different test sets: one of Beatles songs, and one of Bach's music played on a piano. The algorithm performed better on the Bach test set, which consists of very "clean" recordings of a harmonic instrument, whereas the Beatles recordings contain more inharmonic sounds from the drum kit and other percussive sounds. Further studies will investigate the effects of including preprocessing stages such as transient removal (Duxbury, Davies, and Sandier 2001) to isolate the harmonic parts of the signal, along with beat detection (Davies and Plumblev 2004) to give beat-length analysis frames.