イベント
第２回創発システム論セミナー（名古屋大学情報文化学部共催）
2013年6月23日
The 2nd seminar on emergent systems
Acoustical Beamforming and Segmentation for
Bird Vocalization Collection and Processing
Distinguished Professor
Kung Yao
Electrical Engineering Department, UCLA
日時：６月２７日（木）午後4:30 （１時間程度）
場所：情報科学研究科棟第１講義室
Thursday, June 27, 2013
4:30 PM
Information Science Bldg. Lecture room 1
ABSTRACT: In this seminar, we will first present a summary of works of various research collaborators of Prof. C.E. Taylor in a U.S. National Science Foundation grant on “Acoustic Sensor Arrays for Understanding Bird Communication.” Some of these efforts include: collection of very large amount of bird song recordings using advanced acoustical sensor arrays in various natural settings; processing the data using new segmentation and specie classification algorithms for identifying events of interest, and their behaviors; this information/knowledge will then be stored in a large database that will be shared among various research groups; and it will be analyzed by computational-linguistic tools to identify the syntax of the songs and associated linguistic properties.
Next, we will present 2-D and 3-D acoustical beamforming algorithms, their implementation on wireless nodes for field data collections. The beamformer can separates two simultaneous distinct bird vocalizations spatially separated. It uses an iterative algorithm that automatically steers a high gain beam toward the first acoustic source and nearly nulls out the second source and then reverses this procedure. Various simulation and field measured data show the effectiveness of the estimation of the direction-of-arrival (DOA) and localization of the sources, as well as that of source separation. After collection, at the spectrogram level, we use the Bayesian recursive change-point-detection algorithm to estimate the beginning and ending of a bird phrase (for others to do specie classification) by exploiting the fact in the absence of a bird waveform, the background waveform’s entropy is high, while in the presence of a bird signal, the entropy is low.
Demonstration of this segmentation algorithm using field collected bird data will be shown.
お問い合わせ（Contact）：鈴木麗璽(Reiji Suzuki)reiji@nagoya-u.jp
