Recognition of In-Field Frog Chorusing Using Bayesian Nonparametric Microphone Array Processing

In this paper, we exploit Bayesian nonparametric micro-phone array processing (BNP-MAP) for analyzing the spatio-temporal patterns of the frog chorus. Such analysis in real en-vironments is made more difficult due to unpredictable sound sources including calls of various species of animals. An ap-plication of conventional signal processing algorithms has been difficult because these algorithms usually require the number of sound sources in advance. BNP-MAP is developed to cope with auditory uncertainties such as reverberation or unknown number of sounds by using a unified model based on Bayesian nonparametrics. We exploit BNP-MAP for ana-lyzing the sound data of 20 minutes captured by a 7-channel microphone array in a paddy rice field in Oki Island, Japan, and revealed that two individuals of Schlegel’s green tree frog (Rhacophorus schlegelii) called alternately with anti-phase. This result is compared with the video data captured by a video camera with 18 units of sound-imaging devices called Firefly deployed along the bank of the rice field. The audi-tory result provides more detailed patterns of the frog chorus in higher temporal resolutions. This higher resolution enables to analyze fine temporal structures of the frog calls. For ex-ample, BNP-MAP reveals the trill-like calling pattern of R. schlegelii