InsectSound1000 An insect sound dataset for deep learning based acoustic insect recognition

Area of Science:

• Bioacoustics
• Machine Learning
• Ecological Monitoring

Background:

• Automated insect monitoring is crucial for pest and ecological assessments.
• Existing acoustic insect recognition systems require large, high-quality datasets for development.
• The InsectSound1000 dataset addresses this need by providing extensive labeled insect sound recordings.

Purpose of the Study:

• To introduce the InsectSound1000 dataset, a comprehensive collection of insect acoustic recordings.
• To facilitate the training of deep learning models for acoustic insect recognition.
• To support the development of novel digital insect sensors for monitoring applications.

Main Methods:

• Collected over 1000 hours of recordings in an anechoic chamber.
• Utilized a four-channel, low-noise measurement microphone array.
• Extracted and labeled more than 169,000 sound samples from 12 insect species.

Main Results:

• The InsectSound1000 dataset contains over 169,000 labeled sound samples.
• Recordings capture a wide range of insect sound levels, from loud to inaudible to humans.
• Each sample is a high-resolution, four-channel wave file.

Conclusions:

• InsectSound1000 is a valuable resource for training deep learning models for acoustic insect recognition.
• The dataset can accelerate the development of digital insect sensors for automated monitoring.
• The detailed methodology allows for future dataset expansion and adaptation.