SoundChunk: A free open-source and user oriented R package for acoustic data management, sound detection and extraction

In a standard bioacoustic experiment setting, after data collection and before data analysis lies a time-consuming process that consists in extracting and labeling sounds of interest (chunks) from usually long recording files, collected through spreading PAM (Passive Acoustic Monitoring) for instance, and storing them in a structured and exploitable way based on both the meta-data of the initial recordings and the label data. SoundChunk is a user-oriented package in R that aims at providing tools for any R user so that they can go efficiently and comfortably through this process. Usually, the tasks include detecting, labeling, and extracting sound from audio files. With the recent advance of Machine Learning, especially Convolutional Neural Networks, some of these tasks are integrated in the ML framework; however, the learning phase relies on labeled chunks that need to be created. Three types of data are involved in the process: (i) WAVE files, whether they come from the initial recordings or are generated by chunk extraction, (ii) meta-data describing the conditions in which the initial WAVE files were recorded, (iii) label track files, whether they are automatically generated or manually input, that contain timed information about audio files. SoundChunk provides utilities to sanely manipulate each of these three categories and combine them into easy-to-use chunks along with their meta-data. We expose functionalities that: (i) clean and restructure label tracks that are readable by softwares like Audacity, (ii) chop long recordings into small fixed-size slices, (iii) detect chunks both interactively (so that robust detection settings can be found on a subset of the recordings) and automatically (so that the detection is applied across all recordings), (iii) dispatch chunks into structured folder(s), according to their meta-data. MaintainersThe team is open for suggestions and contributions. Contact: soundchunknfeat@proton.me (currently: ASV, PRG). UsersThis document is available as a vignette once the package is loaded an may be used on a set of example data (Villain and Renaud-Goud 2023).