Identifying Actions for Sound Event Classification
Authors:
Benjamin Elizalde,
Radu Revutchi,
Samarjit Das,
Bhiksha Raj,
Ian Lane,
Laurie M. Heller
Abstract:
In Psychology, actions are paramount for humans to identify sound events. In Machine Learning (ML), action recognition achieves high accuracy; however, it has not been asked whether identifying actions can benefit Sound Event Classification (SEC), as opposed to map** the audio directly to a sound event. Therefore, we propose a new Psychology-inspired approach for SEC that includes identification…
▽ More
In Psychology, actions are paramount for humans to identify sound events. In Machine Learning (ML), action recognition achieves high accuracy; however, it has not been asked whether identifying actions can benefit Sound Event Classification (SEC), as opposed to map** the audio directly to a sound event. Therefore, we propose a new Psychology-inspired approach for SEC that includes identification of actions via human listeners. To achieve this goal, we used crowdsourcing to have listeners identify 20 actions that in isolation or in combination may have produced any of the 50 sound events in the well-studied dataset ESC-50. The resulting annotations for each audio recording relate actions to a database of sound events for the first time. The annotations were used to create semantic representations called Action Vectors (AVs). We evaluated SEC by comparing the AVs with two types of audio features -- log-mel spectrograms and state-of-the-art audio embeddings. Because audio features and AVs capture different abstractions of the acoustic content, we combined them and achieved one of the highest reported accuracies (88%).
△ Less
Submitted 5 August, 2021; v1 submitted 26 April, 2021;
originally announced April 2021.
