predict-idlab / gssp_web_app

Web application to acquire picture description speech data according to the GSSP

Speech web app

This repository contains the web app implementation that was utilized to collect data in order to assess the acoustic properties of the Ghent Semi-spontaneous Speech Paradigm (GSSP), a new speech acquisition methodology in which participants were asked to describe images with a constant emotional load.

89 Dutch-speaking participants fulfilled the web-app speech acquisition, which were enrolled through leveraging the researchers' network and the prolific platform.

relevant links:

• The dataset can be found on Kaggle
• a preprint of the manuscript can be found on psyarxiv
• The the GSSP dataset analysis notebooks

Web app structure

Prior to the the web app its data acquisition, the participants loop through the following steps:

1. A Welcome page, which provides a general overview of the study's purpose
2. An Introduction page, which acquired demographics, together with informed consent
   This page also showed guidelines for the GSSP task.
3. The Instructions page, which provides general instructions for the GSSP task. Specifically:
   • three demo videos were shown how the task should be performed
   • the participants were instructed to read aloud the fixed "Marloes" text
4. A 5 minute Rest should bring participants into a neutral baseline state

• The First acquisition consists of the Read-aloud Marloes task, after which the participants fill in their experienced arousal and valence values during the task.
• Afterwards, 5 PSSG Picture descriptions were acquired (alternating between the Radboud and PiSCES image subset). The first shown image always originates from the PiSCES subset. After each image, the participants filled in their experienced arousal and valence values during the task.

This was repeated 6 times, follwed by a Final Marloes acquisition, resulting in a total of 7 Marloes samples, 15 Pisces samples and 15 Radboud samples per participant.

The GSSP is already used in other studies. For example, the fce_stripped branch contains a stripped version of the app in which participants who experienced Adverse Childhood events, filled in this quaestionnaire.

---

Folder structure

└── app
   ├── API                     <- API endpoints / utlities
   ├── static
   │   ├── css
   │   ├── img                 <- images used in the app
   │   │   ├── demo
   │   │   ├── PiSCES
   │   │   └── Radboud
   │   ├── _js                 <- javascript files used in the app (audio recording)
   │   ├── sound               <- sound files used in the app
   │   └── video               <- demo video of GSSP task
   └── templates               <- jinja html templates

---

Running the web app

Via Python

Set first DEPLOY to False in the Appconfig class of app/config.py

# create a virtual environment
virtualenv -p /usr/bin/python3.8 .venv
source .venv/bin/activate

# install the required packages
pip install -r requirements.txt

# start the app
python app/main.py # the app should be accessible on localhost:8080

Via Docker

Make sure that DEPLOY is set to True in the Appconfig class of app/config.py

# build the image 
docker build .
# you should have an output "sucessfully built <IMAGE_ID>" on the last line

# test the image
docker run -it -p 8081:80 <IMAGE_ID>

---

👤 Jonas Van Der Donckt, Mitchel Kappen