Evaluation of an algorithm for the automatic detection of salient frequencies in individual tracks of multi-track musical recordings

This paper evaluates the performance of a salient frequency detection algorithm. The algorithm calculates each FFT bin maximum as the maximum value of that bin across an audio region and identifies the FFT bin maximum peaks with the highest five deemed to be the most salient frequencies. To determine the algorithm's efficacy test subjects were asked to identify the salient frequencies in eighteen audio tracks. These results were compared against the algorithm's results. The algorithm was successful with electric guitars but struggled with other instruments and in detecting secondary salient frequencies. In a second experiment subjects equalised the same audio tracks using the detected peaks as fixed centre frequencies. Subjects were more satisfied than expected when using these frequencies.