Data
This week, I began work on the dataset that will be used with the classification algorithm. I recorded a collection of bird calls at Mansfield Hollow and from my apartment and manually classified them, either by the appearance of the bird or using YouTube. The amount of time and effort that it took to find and classify the birds has proven to us that this would be an inefficient manner of collecting a training dataset. However, it was helpful to get a sense of what kinds of birds we are likely to find in this area and what locations will be good to bring our device in the future once the algorithms are up and running. I made spectrograms out of clips from some of the recordings to start to get a sense of how we should handle our classification (e.g. common frequencies of bird calls in the area, common frequencies of background noise that should be filtered out, etc.) Additionally, these samples will be useful as a testing dataset. In the future, I will work on creating a training dataset using bird call recordings in open source datasets, and I will introduce noise to make the dataset more realistic.