Environment Sound Recognition Systems-A Case Study

Abstract

There are various sounds in the environment, which are produced by multiple sources such as birds, animals, machinery, etc. An Environment sound recognition system (ESRS) detects and categorizes these environmental sounds. ESRS plays a major role in different applications with sound processing such as noise removal, IoT-based systems for sound monitoring, etc. The paper presents simple ESRS models trained on an existing standard environmental sound dataset, freely accessible via the Freesound project. This dataset contains 250000 unlabeled auditory extracts and 2000 short clips representing 50 different basic sound events. Various feature extraction techniques are applied to sound data; the extracted features are then represented using various vectorization techniques so that they can be incorporated into the models. ESRS models are built based on standard Machine Learning(ML) algorithms. Following that, these models are evaluated, tested, and compared.