From f5953b2ffca001c69bac9cfcdc7c028d43a07df4 Mon Sep 17 00:00:00 2001 From: Paul Best <paul.best@lis-lab.fr> Date: Fri, 12 Aug 2022 12:31:52 +0200 Subject: [PATCH] Update README.md --- README.md | 31 +++++++++++++++++++++++-------- 1 file changed, 23 insertions(+), 8 deletions(-) diff --git a/README.md b/README.md index efaddd2..29f678f 100644 --- a/README.md +++ b/README.md @@ -1,14 +1,29 @@ -# carimam_cnn +# CETA CNNs -Convolutionnal neural networks for humpback whale vocalisations and delphinid whistles. +## Project description +This project allows to detect the vocalizations from several cetacean species in acoustic signals using Convolutionnal Neural Networks (CNNs). -Call either run_CNN_HB.py for humpback whales or run_CNN_delphi.py for delphinid whistles. +CNN architectures with pretrained weights are available for inference through a python interface for the following species / signals: +- Humpback whale calls (*Megatera novaeangliae*, trained on recordings from the Caribbean Sea) +- Dolphin whistles (*Delphinid*, trained on recordings from the Caribbean Sea) +- Sperm whale clicks (*Physeter macrocephalus*, trained on recordings from the Mediterranean Sea) +- Fin whale 20Hz pulses (*Balaenoptera physalus*, trained on recordings from the Mediterranean Sea) +- Orcas pulsed calls (*Orcinus orca*, trained on recordings from the North-Eastern Pacific Ocean) -Call the python script followed by the list of wav files you would like to process, and optionnaly an output filename. -For example : -`python run_CNN_HB.py file1.wav file2.wav -outfn predictions.pkl` +For the detection of Antarctic fin whale and and Antarctic blue whale vocalizations, see https://gitlab.lis-lab.fr/paul.best/at_bluefin -This script relies on torch, pandas, numpy, scipy, and tqdm to run. Install dependencies with pip or conda. +## Usage +Use python to run the script forward_CNN.py along with a target specie and a folder of audio files to analyse. A tabular file will be saved with the model's predictions for the corresponding signal to detect (probability of presence). + +Run `python run_CNN.py -h` for a detailled API. + +The output file with predictions can be read in python using pandas : `pandas.read_pickle('filename.pkl')` + +## Dependencies +This script relies on `torch`, `pandas`, `numpy`, `scipy`, `soundfile` and `tqdm` to run. You can install them using pip or conda. If a GPU and cuda are available on the current machine, processes will run on GPU for faster computation. -paul.best@univ-tln.fr for more information + +## Contact +You can reach me at paul.best@univ-tln.fr for more information + -- GitLab