Bird Sounds: Classification and Visualization

As a part of the Google’s “A.I. Experiments” (https://aiexperiments.withgoogle.com/) this project was developed in conjuction with the Cornell Lab of Ornithology. The main purpose of this was to create an AI for classifying and visualizating the sounds made by different species of birds and group them by similarities.

For doing this, Deep Learning techniques were used, but, first the sounds needed to be fragmented into “elemental” divisions, such as when analyzing voice. Then, it was needed to create fingertips from this “elemental” bird sound/particle for classifying new sounds and grouped them by similarities. For this last part, a machine learning technique known as t-distributed stochastic neighbor embedding, or t-SNE (https://en.wikipedia.org/wiki/T-distributed_stochastic_neighbor_embedding), this models each high-dimensional object by a two- or three-dimensional point in such a way that similar objects are modeled by nearby points and dissimilar objects are modeled by distant points, this is particularly useful comparing the different elemental footprints that can create different bird sounds.

After this, the AI now can identify the variations in these sounds and to determine if it is another kind of bird or the same, also it can name the bird by only hearing the sound it emits.

The ultimate goal of the project is to introduce microphones in forests and jungles in order to get more information, classify the noises automatically and to monitor the different species.

For more information: https://aiexperiments.withgoogle.com/bird-sounds