City Classification from Multiple Real-World Sound Scenes

Abstract

The majority of sound scene analysis work focuses on one of two clearly defined tasks: acoustic scene classification or sound event detection. Whilst this separation of tasks is useful for problem definition, they inherently ignore some subtleties of the real-world, in particular how humans vary in how they describe a scene. Some will describe the weather and features within it, others will use a holistic descriptor like 'park', and others still will use unique identifiers such as cities or names. In this paper, we undertake the task of automatic city classification to ask whether we can recognize a city from a set of sound scenes? In this problem each city has recordings from multiple scenes. We test a series of methods for this novel task and show that a simple convolutional neural network (CNN) can achieve accuracy of 50%. This is less than the acoustic scene classification task baseline in the DCASE 2018 ASC challenge on the same data. A simple adaptation to the class labels of pairing city labels with grouped scenes, accuracy increases to 52%, closer to the simpler scene classification task. Finally we also formulate the problem in a multi-task learning framework and achieve an accuracy of 56%, outperforming the aforementioned approaches.

An evaluation of data augmentation methods for sound scene geotagging

“…Sound scene geotagging is a relatively novel topic of research. It means to correctly identify the geographical position where a recording was captured [1] and as such it can also be described as either audio geotagging or as a more specific task such as audio-based city classification. Subject to the precision of the dataset annotations, it could be formulated as a coarse target problem (e.g.…”

“…Sound scene geotagging is a task with many future realworld applications, such as automatically knowing a precise geo-location from the background sounds of an emergency call that could potentially decrease the response time [2] or it could position a person in a location with no cameras at a specific time for criminal investigations. However, the current state-of-theart in audio geotagging that uses multi-task learning, achieves only 56% accuracy [1] on city classification.…”

“…Prior to the work in [1], there was little investigation of audio geotagging, and to the best of our knowledge, not as a classification task. Examples by Elizalde et al [6] and Kumar et al [7] used sound scenes generated with sound events not specific to the locations.…”

“…In [1], the authors use a dataset labelled for both audio geotagging (in a city level) and acoustic scene classification. As such, the annotations function as a many-to-many connection between the city and scene classes.…”

An Evaluation of Data Augmentation Methods for Sound Scene Geotagging

“…City classification accuracy by data augmentation method with different CNN models.Accuracy (%) obtained with the multi-task CNN from[1] cyclic drop stretch cyclic+stretch all benchmark…”