Influence of the external ear of Hipposideros pratti on determining the direction of constant frequency sound source using Residual Networks

To explore the relationship between the external ear and spatial localization in bats using constant frequency signals, the influence of bats’ external ear on determining the direction of constant frequency sound source was analyzed using deep learning algorithm with the help of static biomimetic binaural receiver. Firstly, a biomimetic binaural receiver with different angles and distances between two ears was designed based on the ear model of Hipposideros pratti. Constant frequency signals with different frequencies were collected from multiple spatial directions. Spectrograms of the synchronized signals received by two ears were extracted and normalized as input features. Finally, the spatial direction of the sound source was estimated using Residual Networks. Experimental results show that the average orientation error of the static binaural receiver to the constant frequency source is basically less than 3.5°, which is higher than that of dynamic monaural receiver. The orientation accuracy is related to the frequency and spatial direction of the sound source. The accuracy is relative high when the sound source is within ±30° along the horizontal direction of the receiver. Angles and distances between ears could also influence the accuracy, and the former’s effect is more significant.