Music Chords Identification From Audio Signals

Abstract

This thesis describes a template based music chord generating technique. It requires only a single instrumental melody file and provides you a chord sheet. In order to achieve the goal, author used digital audio processing, plus some essential theories of music. The audio file is processed and necessary information like beat, frequencies and some timing related facts are grabbed. In parallel, chord classification process is done, in order to narrow down the possibilities of chord which can occur for a given time. Chord classification is a onetime process and results can be used for any song. Past research can be found in the same domain of computerized music chord abstraction. But most of them are on chord detection, not chord prediction. Chord detection is something like detecting existing chord in audio, and chord prediction is proposing suitable chord according to the melody. However the methodologies and findings can be taken into this subdomain as well. Research exists to predict music chords, but they only deal with Jazz music chord prediction where repetitive chord occurring patterns exist with slight differences. Author has shown a systematic way of finding chords for songs which do not have repetitive chord blocks. This research is divided into two sections in order make it clearer. First section is on audio file analysis, which uses MATLAB for audio signal analyzing. The second section is on process notation and predicting relevant musical chords, which uses Java for implementation. Final system is not bundled into one product as a whole, but comes with two separate modules which carry out their own tasks independently.