A micro-control device of soundscape collection for mixed frog call recognition

Abstract

An Ecological Forest Park is a place that combines leisure and research, but the balance of the local ecology can be affected if the number of tourists exceeds the quota allowed for the park. Ecologists have utilized wild soundscapes in the most common surveys of frog ecology. However, soundscapes for a wild field are highly complex when recorded into a single channel from multiple sources since it contains various types of background voices and an unknown number of mixed sources. Blind source separation is ineffective in later processing of voiceprint recognition algorithms. This paper uses a micro server for automatic directional control of the microphone facing the animal source. This device also uses an interference tube to eliminate the noise outside from the directional microphone to predict the number of mixed sources that are used for the blind source separation by the cluster of frogs and discrepancy in the croaking gap. In the end, adaptive multi-stages average spectrum (AMSAS) is used to separate the animal sources, and the experiment makes use of the recorded files including the monosyllables of six types of frogs and mixed ones with seven kinds from the Shan-PING Forest Ecological Garden. Meanwhile, we compare the recognition rates among the processing using dynamic time warping, multi-stage average spectrum, and AMSAS to verify the superiority of the proposed method.