The goal of this deep learning model is to identify the context in which a cat meows. For this purpose, we use a dataset of 440 meows from 21 cats (two breeds: Maine Coon and European Shorthair), recorded under three eliciting contexts: Brushing, Isolation in an unfamiliar environment, and Waiting for food. Each stimulus lasted up to 5 minutes and recordings were collected in naturalistic settings with standardized handling to minimize stress. This dataset is provided by the university of Milan at this address : CatMeows: A Publicly-Available Dataset of Cat Vocalizations. We also based our research on this paper : Audio Deep Learning : Sound Classification
Audio labeling is a method widely used in machine learning, agronomy, sociology, music, etc. We therefore propose to present an approach for processing audio signals and labeling these signals.
Our example will be cat audio labeling. The dataset contains 440 audio files in .WAV format. Each audio file is a cat meow. The cats were recorded in different situations (being brushed, waiting for food, isolated in an unfamiliar space). The goal here is to determine the emission context of a meow.
- Preprocessing of audio data
- Feature extraction via a neural network and convolutional layers
Project by Augustin Antier, Lena Causeur, Louis Prusiewicz-Blondin for the Third edition of the Autumn School on "Machine Learning in the Life Sciences" by the Institut Agro Rennes-Angers