Musical Instrument Recognition Based on the Bionic Auditory Model

We present a bionic auditory system for musical instrument recognition. This system is designed based on the physiological structures of the human auditory system that are essential to sound source recognition, such as the basilar membrane and inner hair cells in the cochlea of the inner ear, cochlear nucleus, and the auditory cortex. A large solo database consisting of 243 acoustic and synthetic solo tones over the full pitch ranges of seven different instruments (guitar, harp, horn, piano, saxophone, trumpet, and violin) is used to encompass different sound possibilities of each instrument. The gamma tone model, the Meddis model, and poster ventral cochlear nucleus (PVCN) model are constructed to imitate the basilar membrane, the inner hair cells, and the cochlear nucleus, respectively. By using 33%/67% splits between training and test data, a self-organizing mapping neural network (SOMNN) based on the function of auditory cortex is established to classify the instruments. The instruments are recognized with an overall success rate of over 75%. This bionic auditory system indicates high efficiency and high accuracy in musical instrument recognition.