Deductive Refinement of Species Labelling in Weakly Labelled Birdsong Recordings

TL;DR

This paper presents a two-step method for precise bird vocalization classification in recordings, improving label accuracy using weakly labeled data, with significant performance gains demonstrated on synthetic datasets.

Contribution

It introduces a novel two-step approach combining detection and classification that enhances species labeling accuracy in weakly labeled bird recordings.

Findings

Achieves 61% accuracy on synthetic dataset.

Reaches 89% accuracy on larger vocalizations.

Demonstrates effectiveness of the method in controlled experiments.

Abstract

Many approaches have been used in bird species classification from their sound in order to provide labels for the whole of a recording. However, a more precise classification of each bird vocalization would be of great importance to the use and management of sound archives and bird monitoring. In this work, we introduce a technique that using a two step process can first automatically detect all bird vocalizations and then, with the use of 'weakly' labelled recordings, classify them. Evaluations of our proposed method show that it achieves a correct classification of 61% when used in a synthetic dataset, and up to 89% when the synthetic dataset only consists of vocalizations larger than 1000 pixels.