AES E-Library

Automatic Soundscape Classification via Comparative Psychometrics and Machine Learning

Computational acoustical ecology is a relatively new field in which long-term environmental recordings are mined for meaningful data. Humans quite naturally and automatically associate environmental sounds with emotions and can easily identify the components of a soundscape. However, equipping a computer to accurately and automatically rate unknown environmental recordings along subjective psychoacoustic di-mensions, let alone report the environment (e.g., beach, barnyard, home kitchen, research lab, etc.) in which the environmental recordings were made with a high degree of accuracy is quite difficult. We present here a robust algorithm for automatic soundscape classification in which both psychometric data and computed audio features are compared and used to train a Naive Bayesian classifier. An algorithm for classifying the type of soundscape across different categories was developed. In a pilot test, automatic classification accuracy of 88% was achieved on 20 soundscapes, and the classifier was able to outperform human ratings in some tests. In a second test, classification accuracy of 95% was achieved on 30 soundscapes.

Authors: Rajagopal, Krithika; Minnick, Phil; Leider, Colby
Affiliations: University of Miami, Coral Gables, FL, USA; audio Precision, Beaverton, OR, USA(See document for exact affiliation information.)
AES Convention:131 (October 2011) Paper Number:8581 Import into BibTeX
Subject:Auditory Perception

Click to purchase paper or login as an AES member. If your company or school subscribes to the E-Library then switch to the institutional version. If you are not an AES member and would like to subscribe to the E-Library then Join the AES!

This paper costs $20 for non-members, $5 for AES members and is free for E-Library subscribers.

Learn more about the AES E-Library

E-Library Location: (CD 131Papers)   /131/8581.pdf

Start a discussion about this paper!