Koogu

A python package for developing and using Machine Learning (ML) solutions in Animal Bioacoustics.

Koogu (ಕೂಗು) is a word in the Kannada language and means "call" (used as a noun) or "to call" (used as a verb). The package offers tools for -

• preparing and processing audio for training ML models,
• training ML models, and
• using trained ML models for automatic recognition.

Installation

The package is available on PyPI, and can be installed as follows

pip install koogu

Dependencies

TensorFlow

koogu uses TensorFlow as backend ML framework. Please ensure that either a CPU or a GPU version of TensorFlow is installed prior to installing koogu.

librosa

koogu uses librosa for reading audio files (only). Please refer to its GitHub page for details about its dependencies and how to they may be addressed.

Others

Besides TensorFlow, all other dependencies will be automatically installed.

How to use Koogu

Koogu offers tools for ML development from the simplest of bioacoustics applications to more complex scenarios. All stages of the workflow ( preparation, training, inference) can be performed independently.

For simpler forms of supervised learning, data must be available as a collection of audio files along with the necessary annotations. The annotations must be in Raven Lite / RavenPro selection table format, which is basically a simple tab-delimited text file providing (at a minimum) the start-time, end-time and label for each event/call. The function from_selection_table_map() in the module prepare_data helps prepare the data, with full customization support, for training a model. The data.feeder module provides classes for efficiently feeding prepared data into a training/evaluation pipeline while the module train_and_eval provides a convenient and customizable interface for performing the actual training.