Tutorial: Ketos Commands

This is a CLI ( command-line interface) tool that offers a set of convenient executables for streamlining the development of acoustic deep learning models for detecting marine mammals.

Currently, there are two commands available: ketos-run and ketos-metrics. Ketos-run allows you to execute models on a directory filled with audio files, offering multiple options for running these models to match your specific needs. Ketos-metrics follows directly afterward and allows you to pass the detections outputed by the ketos-run command and a set of annotations and compute performance metrics.

However, we are actively expanding our toolset and plan to introduce the following commands to make Ketos even more robust:

  1. Data Preparation: Create training and test datasets from raw acoustic data by converting the audio into different formats such as spectrograms, cepstrogram, etc.

  2. Model Training: Train a deep learning model to not only detect but also classify NARW vocalizations. With this, you can easily distinguish different types of vocalizations within the acoustic data.

  3. Model Evaluation: Evaluate your trained model on fully-annotated test datasets. This command will also generate a report containing standard performance metrics to help you understand the effectiveness of your model.