Acoustic Scene Classification: Classifying environments from the sounds they produce

In this article we present an account of the state-of-the-art in acoustic classification (ASC), the task of classifying environments from the they produce. Starting from a historical review of previous research in area, we define a general framework for ASC and present different imple- of its components. We then describe a range of different algorithms for a data challenge that was held to provide a general and fair for ASC techniques. The dataset recorded for this purpose is, along with the performance metrics that are used to evaluate the and statistical significance tests to compare the submitted methods. use a baseline method that employs MFCCS, GMMS and a maximum likelihood as a benchmark, and only find sufficient evidence to conclude that algorithms significantly outperform it. We also evaluate the human accuracy in performing a similar classification task. The best algorithm achieves a mean accuracy that matches the median accuracy by humans, and common pairs of classes are misclassified by both and humans. However, all acoustic scenes are correctly classified by least some individuals, while there are scenes that are misclassified by all.