A pipeline to create, upload and analyse long format recordings using the Zooniverse citizen science platform.

We have an open project aimed at adding vocal maturity labels to segments LENA labeled as being key child in Zooniverse (https://www.zooniverse.org/projects/chiarasemenzin/maturity-of-baby-sounds).

If you would like your data labeled, here is what you'd need to do.

1. Get in touch with us, so we know you are interested! (authors' contact information at the bottom of this README)
2. Have someone trustworthy & with some coding skills (henceforth, the RA) process your LENA data using these instructions.
3. When you are done, this will have generated a meta-data file, containing the linkage between the original .its and the ~250 500-ms clips/child. Keep the metadata safe - without it, you won't be able to interpret your results!
4. Have the RA create an account on Zooniverse (top right of zooniverse.org) for them and yourself, & provide us with both handles. This person should first update the team section to add you (have ready a picture and a blurb). They can also add your institution's logo if you'd like. Both of these are done in the lab section
5. They will then follow the instructions in this README to create subjects and push up your data -- see below.
6. We also ask the RA to pitch in and help answer questions in the forum, at least one comment a day.
7. You can visit the stats section to look at how many annotations are being done. Since the process of data generation takes up Zooniverse resources, we'll manage requests and try to not make very many in succession. Please do NOT ask for data to be generated.
8. When we get data back (every month or so), we'll update it in an OSF repo and let you know. We provide code to derive key analyses, which involve using your meta-data to remove data belonging to other people.

This project comprises three folders:

• /create_subjects contains scripts to segment daylong recordings into small chunks (500ms) and write a metadata file.
• /upload_subjects contains a script to convert and upload data on the zooniverse platform, using the Panoptes API.
• /zoon_analysis contains scripts to convert Zooniverse data to json format and perform analysis in R.

This README assumes you know your way around a Terminal. If you don't, follow a tutorial, e.g. https://swcarpentry.github.io/shell-novice/

Minimum requirements: 1.6 gHz intel core (dual) 4GB 1600 MHz DDR3; 30GB of memory

• Python 3.6 or later required.

It is recommended that you run this software within a virtual environment. To set up a Python3 virtualenv follow these steps:

Find here instructions on how to install Python for Linux , MacOS and Windows.

For example to create one in the local directory called ‘mypython’, type the following:

Mac OS / Linux

python3 -m venv mypython

Windows

py -m venv mypython

You can activate the python environment by running the following command:

Mac OS / Linux

source mypython/bin/activate

Windows

mypthon\Scripts\activate

Then you can confirm you’re in the virtual environment by checking the location of your Python interpreter, it should point to the env directory.

On macOS and Linux:

which python
.../env/bin/python

On Windows:

where python
.../env/bin/python.exe

As long as your virtual environment is activated pip will install packages into that specific environment and you’ll be able to import and use packages in your Python application.

If you want to switch projects or otherwise leave your virtual environment, simply run:

deactivate

If you want to re-enter the virtual environment just follow the same instructions above about activating a virtual environment. There’s no need to re-create the virtual environment.

Python packages required:

• lxml
• pandas
• pydub
• panoptescli

all packages can be installed with pip:

pip install pydub

Clone or download the repository on your local machine by running:

$ git clone https://github.com/psilonpneuma/Zooniverse.git

To check Python, FFMPEG and required packages installation status run:

$ cd Zooniverse
$ python installation_check.py

Open the configuration file config.py that you will find at the top level of this repository in any text editor (e.g., Sublime Text), and specify the local paths:

• "python": python interpreter. You can find which version of python you're running by typing python -V on your terminal. If you only have one version of python and it's Python 3, you can write "python":"python". If you have several, including python3, you may be able to call it with "python":"python3". If using a virtual env, enter the path to the environment Python executable (e.g. /home/attie/projects/thing/venv/bin/python3).
• "infolder": the folder with its and wav files. Make sure corresponding its and wav files are named in the same way, e.g. if the .its file is called e1234.its, the .wav file should be called e1234.wav
• "outfolder": this is the folder where you want this script to extract final clips and the metadata.
• "metadata_fn": name of the metadata file; choose anything you'd like since it will be created in the process. We recommend naming it as follows: metadata_LABNAME_DATEISO_INITIALSUPLOADER1 where LABNAME is a short name identifying your lab (it can also be a random name, if you want to mask your data's identity, eg if you only work with one infant population); DATEISO the date in ISO (YYYYMMDD), INITIALSUPLOADER the initials of the person uploading, and a digit.
• "batch_name": name that will be used to label the batch during uploading, choose anything you want as long as there are no spaces and it is unique (i.e., it will not have been used by you or any others in the past). NOTE! You cannot use the same name twice (i.e., for two subsequent uploads). We recommend naming it as follows: LABNAME_DATEISO_INITIALSUPLOADER1 where LABNAME is a short name identifying your lab (it can also be a random name, if you want to mask your data's identity, eg if you only work with one infant population); DATEISO the date in ISO (YYYYMMDD), INITIALSUPLOADER the initials of the person uploading, and a digit.

See the configuration file for examples. Don't forget to put a "/" at the end of each path, and make sure the path does not include spaces.

We foresee three phases and explain each in turn.

First, short clips need to be created from the long recording files.

You may need to navigate to the relevant folder by doing:

cd Zooniverse/create_subjects/

Then simply run a command like the following to get started:

python pipeline.py

At the location specified in outfolder in the config.py file, the software will create:

• an intermediate/ folder with extracted CHN chunks
• a data-for-upload/ folder with the final short clips and a metadata file

Warning: Although there is in principle no limit to the number of long files that can be processed in the first stage, the Zooniverse platform guidelines recommend uploading 1000 files or less at a time.

For this reason, the creation of clips and their upload are kept here as two separate stages. As a result, you might need to run the steps in this section more than once, separating your clips into batches of 1,000 files, and changing the batch name in your config.py file each time.

First, navigate to the relevant folder by doing:

cd Zooniverse/upload_subjects/

Or, if you just created your subjects with the instructions above, then do:

cd ../upload_subjects/

Next convert your clips to mp3 format by running:

$ ./convert_2_mp3.sh

if you encounter a permission error, run:

$ chmod +r convert_2_mp3.sh
$ ./convert_2_mp3.sh

If that still doesn't work, try

$ bash convert_2_mp3.sh

Once your files are in mp3 format, check how many you have in the folder. For instance, you can use ls YOUR_OUT_DIR | wc -l (making sure to replace YOUR_OUT_DIR with the actual path of the folder where you're staging your uploads). If there are more than 1k files, split them up by creating new directories (e.g., mkdir YOUR_OUT_DIR2) and moving excess files there until there are 500-1000 files in each. Since they are numbered randomly, you can split them by the first digit. For instance, if you have 2k files initially, you can do mv YOUR_OUT_DIR/[45689]*.wav YOUR_OUT_DIR2 to move half of the files to the new directory.

To begin the upload, first of all, configure and save your default Zooniverse login details by running:

$ panoptes configure

You should see the following output:

endopoint[www.zooniverse.com]
username []: 
password []: 

The endpoint value can be left unchanged, just press Enter. Type in your zooniverse username and password. The API will store this information and you will not need to enter it again.

Next, for the first directory with files to upload, run:

$ python upload_data.py

There will be some text written out, and eventually, the script will pause and you will see something like what follows:

$ python upload_data.py 
Started.
Your settings:
/Users/acristia/Documents/Zooniverse-data/LAAC_20200418_ac1_for_upload/

The script will then access the batch folders created in Step 1 (create_subjects), using the outfolder path specified in the configuration file. It will automatically access the subfolder for_upload and loop through the batches, uploading them one at a time. As it does so, you will see the batch name being processed, and the following progress bar:

Uploading subjects  [####################################]  100%          
Uploading subjects

Important When the upload is complete, go to the (Workflow)[https://www.zooniverse.org/lab/10073/workflows/12193] section of the Project Builder in Zooniverse (you need to be logged in for the direct link to work!), and tick the name(s) corresponding to your new batch(es) to add it to the current workflow.

At this point you are done! Celebrate as you wish.

• Analysis

(This section has not yet been beta-tested.)

After having downloaded the CSV file from the shared OSF repository, launch the notebook Convert_Zooniverse.ipynb to obtain a classification files of type:

Idx, UserID, AudioData, Answer, Dataset, Question

• Recovering your audio files from the metadata and original WAV and ITS files.

To recover the anonymised clips created, in case of loss or deletion, you can run recover_chunks.py.

The script requires:

• A metadata file, as generated from pipeline.py
• The WAV files
• The (corresponding) ITS files

Usage:

$ python3 -i /path/to/input/folder -o path/to/output/folder -md /path/to/metadata

The input folder is where your WAV and ITS files are stored, and the output folder is where the recovered clips will be created.

Chiara Semenzin [email protected] Alex Cristia [email protected]