”Harmonic-Mixing” is a DJ technique, which enables the DJ to mix songs together based on musical keys. In order to make it achievable, the keys of the songs must be known. The aim of the thesis is to discover the most suitable keydetection-algorithms to create a DJ-assistant software in the future. The first part of the thesis describes known key detection techniques and algorithms, then it demonstrates four methods implemented by me.
The basis of the program is a framework in MATLAB environment, whitch can analyse a huge song database with the created easily customizable algorithms. The results of the measurements are exported in LATEX format. The framework creates clear reports and a summary of the whole analyzing process, the results of the keydetections are also included in these documents. The free parameters of the algorithms can be tuned easily, based on the exported results. The algorithms are using spectral-analysis to detect the musical keys of the songs. The implemented methods are: musical note intensity extraction from audio signals by energy spectrum density analysis; collection of the most intensive notes from windowed signals; chord-detection and the summation of the identified chord notes; and determination the local maximums of the trend compensated amplitude-spectrum.
With the optimal choice of parameters, meeting certain cicumstances, the methods became nearly the same accurate as a well-known keydetection software (in case of my song database). The thesis includes the method of the free parameter tuning, and covers the future prospects of the program.