Quickstart: CLI Workflow

Here’s the complete sequence of commands for the NARW project:

The example uses a light-weight custom DenseNet architecture.

Explore the dataset (optional)

$> koogu-explore narw_project.config

1. Data preparation

With audio-annotation mappings and data transformation settings already defined in the config file, you can simply run the below command to prepare the data for training.

$> koogu-prepare narw_project.config

2. Training

With network architecture and training/hyperparameter settings already defined in the config file, you can simply run the below command to start training. Specify a name for the model so you are able to distinguish models from different training attempts.

$> koogu-train narw_project.config narw_detector_v1

3. Performance assessment

3.1. Run on test dataset

# 4. Assess performance

$> koogu-assess narw_project.config narw_detector_v1_perf.html

# 5. Process field recordings

$> koogu-recognize narw_detector_v1/ field_audio/ detections/ \
     --threshold 0.75 --recursive --reject-class Other

# 6. Export for Raven Pro

$> koogu-export narw_project.config raven --model narw_detector_v1 \
     --threshold 0.85 --suppress Other