This recognizer is designed to detect Ovenbird (Seiurus aurocapilla) songs in audio recordings. The recognizer output is a sigmoidal score for Ovenbird presence and the relative decibel level for each ~0.275 second detection window of a recording. This notebook also includes a model that uses the relative decible level of Ovenbird detections to estimate the distance of the Ovenbird from the ARU with upper and lower bounds of a 95% confidence interval. Please read the detailed description of the limitations of the distance estimation model in the Ovenbird_Recognizer_Processing Jupyter notebook.
Recordings must be run through the Ovenbird detection model by running the scripts in the Ovenbird_Recognizer_Processing Jupyter notebook either locally or via Google Colab. The repository can be cloned to a local machine in order to run the recognizer via Jupyter Notebooks, or to Google Drive where the recognizer can be run in the Google Colab environment. To run the notebook on Google Colab you will need to have a Google account and have the audio recordings that you wish to process stored in your Google Drive. If your recordings are stored locally on your machine or a portable hard drive you will need to run the code locally on your machine within a python environment that runs Jupyter Notebooks.
If you wish to run the model on your local machine you can clone the repository to your machine, or you can download all three parts of the model (oven-model.index, oven-model.meta, and oven-model.data-00000-of-00001) and the notebook (Ovenbird_Recognizer_Processing.ipynb) to the same folder on your local drive and run the notebook in a local python environment of your choice. You'll need to have the model files and notebook in the same folder that you will set to be your working directory, though the recordings can be located elsewhere on your machine.
If you want to run the model using Google Colab rather than a python environment on your own computer then you can open the notebook directly in Colab from the Github repository and then clone the repository to your Google Drive via the steps in the notebook. Follow the steps for linking your Google Drive account and the audio recording files stored there to the scripts in the code in make them accessible to the model running in the notebook.
This recognizer was developed using the Tensorflow package in python by Chris Scott in collaboration with Elly Knight.
The distance estimate model was developed by Dan Yip and Elly Knight.
Kevin Kelly developed these notebooks and updated the scripts