M4sum/save-forest-elephants

Detect elephant rumbles and gunshots on recordings made in the forests of central Africa and optimize the prediction process to implement on-edge. README IMPORTANT NOTE: This is still a work in progress. the current best working solution is published under the branch 'integration_new_model'. To replicate the solution, you need to follow the readme present under that branch. This is the README for the 'Inference_pipeline.py' script used for processing data and running the 2-stage model for elephant call prediction. This script runs the inferance prediction pipeline. The prediction pipeline consists of 2 main components: 1) Data processing. Given a directory with .wav files, generate the corresponding spectrogram representations. 2) Generate 2-stage model predictions: 2a) Generate 2-stage model segmentation. Predict the 0/1 time segmentation given the 2-stage model. 2b) Elephant call prediction. Output the models predictions for start / end times of calls in csv formatt. The script allows for complete and partial runs of the 2 steps above: - Flag '--process_data': Process the '.wav' files (step 1). - Flag '--make_predictions': Generate elephant call predictions (step 2). For each part of the pipeline, command line arguments are used to properly specify necessary data. KEY FILES This folder contains all of the python scripts needed to run the primary script 'Inference_pipeline.py'. Additionally, the folder '2_Stage_Model/' contains the components of the 2-stage model. When you run the script, unless otherwise specified, the default location for outputed values are as follows: - Processed Spectrograms: By default a folder 'Spectrograms/' will be created in this directory containing the processed spectrograms. Specifically, these processed spectrograms will exist in a sub-directory of 'Spectrograms/' with name based on the folder containing the original '.wav' files. To change this default data location use the flag '--spect_out' - Model 0/1 segmentations: You will not need to use this data! This is just used for the model to save its segmentations before post processing to get the actual start/end t …