Invasive Species Detection
Open-source audio analysis tool designed to detect rare audio features within field recordings.
Upload raw field data and a sample of the target feature. The tool will automatically identify features of interest and apply labels to segments.
Use grid view mode to quickly efficiently annotate high-priority samples and fine-tune a classification model that can be exported for use in the field.
Point map mode provides a convenient method of visuallising large quantities of audio data. It is helpful for identifying outliers and incorrectly labelled features.
This audio analysis tool uses a method called active few-shot learning. A pre-trained model (Audio Spectrogram Transformer) is used to generate embedding vectors of uploaded features and prototypical few-shot learning provides data-efficient inference.
Learn more about this method here.
