Modeling a Melody Recognition Task using a Cohort Network

Dalla Bella, Peretz, and Aronoff studied the effects of musical familiarity on melody recognition by comparing performance between musicians and nonmusicians in a melody gated-presentation (MGP) task. They identified three events in this task which were the familiarity emergence point (FEP), the isolation point (IP), and the recognition point (RP). The FEP occurred earlier in musicians than nonmusicians, but the IP occurred earlier in nonmusicians. Finally, the RP occurred slightly earlier in musicians. We simulated the qualitative results of the MGP task using a connectionist simulation of the cognitive processes underlying the emergence of these three events. We call this a melody cohort network (MCN). Separate neural networks modeled musicians and nonmusicians where the musician network represented a larger corpus of stored melodies. The MCN consisted of a core network which modeled the IP and meta-level networks which used the core network as input to model the FEP and RP. Our MCN captures the qualitative results of the MGP task and shows how stored memory size may affect the melody recognition process. We also used the simulation to predict the effects of two levels of severity of acquired amusia characterized by elevated thresholds for perceiving changes in pitch.