Musician style recognition: Training a computer system to recognize the signature patterns of pianists

Every musician has his/her own style of playing music. Similar to handwriting, each musician's rendition of a song differs somewhat from that of others. This research limited its focus to the stylistic patterns found in jazz and blues piano playing. Several factors were identified as contributing to the unique “style” of a pianist. These factors include: keyboard range, note density, rhythm lengths, pedal usage, dynamic range and tempo variance. A program was implemented to analyze music from MIDI files. Metrics were developed to quantify the important factors. These metrics were then used to capture the style for five different performers. The results of the analysis during the “learning phase” were then subsequently used during a “testing phase.” The testing phase introduced new songs performed by one of the five pianists. The program developed as part of this research was used to identify the performer of the new song. Two strategies of identification were tested. One identified the performer based on the learning-phase song that was closest to the new song. The second strategy created a “composite” for each performer during the learning-phase and then used those composites to identify the performer. The program was tested using both strategies to determine its ability to distinguish between the pianists. The details of the metrics used and the results of the tests conducted are discussed.