Real-time system for bird sound recognition

Environmental sound processing is one of the main research areas in biodiversity preservation studies. One of the most important components of these studies is the sound processing system to be used. In this study, we propose a stand-alone, low-level, custommade, real-time environmental sound processing system concentrated on single-labeled bird calls and composed of a microphone circuitry, a Texas Instruments Tiva C Connected Launchpad (consisting of an ARM Cortex-M4F based microcontroller), and a storage unit. The proposed system enables data recording and on-board preliminary signal processing, feature extraction, classification and data storage. In the proposed system, we simultaneously record and process data. As the first processing step, we filter out steady background noise using spectral noise gating technique. Secondly, we detect necessary sound signal parts. We then extract mel frequency cepstrum coefficients (MFCCs) as features from the sound signal and classify features by minimum distance classifier with trained features. As the last step, store class labels after classification on an SD card. This is done by effectively defining and utilizing a ping-pong buffer structure on the microcontroller. The proposed system offers flexibility (both in hardware and software) for expansion.