Describing vocalizations in young children: A big data approach through citizen science annotation
This repo contains the code needed to reproduce a conference paper (202007_SLT) and a journal paper (202010_jslhr). Given that the latter is a later-produced expansion of data and analyses, we only provide information for reproducing the latter.
To reproduce the manuscript, you can simply knit 202010_jslhr/paper.Rmd. NOTE! If you are not in the LAAC team, this is probably the only step you can reproduce. If you want to reproduce the other steps, write to alecristia@gmail.com
To rerun the whole pipeline:
- Put the following two files inside files_from_zooniverse/
- maturity-of-baby-sounds-subjects.csv contains information about all the clips that have been put up on zooniverse
- maturity-of-baby-sounds-classifications.csv contains information about all the classifications that have been done on zooniverse
These are very large, so they are not synced in the repo.
- Run data_analyses/code/generate_jslhr_data.R. This will just select down the subject and classification data to the filenames in the PU dataset. In addition, this step will generate
key_info.csv, which is used in paper.Rmd, as well as the following files, which are used in preprocess.R:
- zoo_subj_info_pu.csv contains information about the PU clips that have been put up on zooniverse
- zoo_anno_info_pu.csv contains information about the classifications that have been done on zooniverse for the PU clips specifically
- Run 202010_jslhr/postprocess.R, which has three steps: cleaning errors in the classifications, generating chunk- and segment-level data. NOTE!! that this file generates
data_analyses/output/clean_classifications.csv, another large file that is not synced but is necessary for this process. It'll generate the two key files which are used in paper.Rmd:
- zoo_lab_maj_judgments.csv
- chunks_maj_judgments.csv
- knit 202010_jslhr/paper.Rmd