mirror of
https://github.com/Bloktard/workshop_stressometer.git
synced 2025-10-29 02:06:01 +00:00
43 lines
2.3 KiB
Markdown
43 lines
2.3 KiB
Markdown
# workshop_stressometer
|
|
Using and creating arduino Nano 33 BLE sense rev2 to train own ai-recognetion model
|
|
|
|
We worked in a team of two between 42 student and art school student.
|
|
|
|
The goal of this workshop was : AI - Arduino - Psychological issues
|
|
|
|
Overall, the goal was to reflect on what is the norme, what is the norm in an AI.
|
|
|
|
Our approach is based in the childhood, where most of the trauma begin.
|
|
In order to ilustrate the trauma using arduino and an AI, we decided to base on reasoning on "traumatizing" a child.
|
|
|
|
So, we trained an AI using Edge-Impulse to recognize some predetermined sentences. The library is in the github (the sentences are in french).
|
|
|
|
We portrayed the trauma using a stress variable. This variable is influenced by different factors :
|
|
|
|
- The mouvment of the arduino
|
|
- The proximity of someting near the arduino
|
|
- The level of sound the arduino hears
|
|
- The recognetion of the sentences
|
|
|
|
In order to show the trauma, the stress variable move up and down and influence a base-stress variable that start at 0. Once the base-stress variable moves up, the only way to lower it is to appease the "child".
|
|
|
|
How we trained the ai-model :
|
|
|
|
We make lot of record on Edge Impusle with different label like "Stress" and "calm" in data acquisition. After that we create an Impulse design with MFCC block audioand and add a classifier, save your impusle.
|
|
Now you can test or train every category.If you are fine go to deployment and choose your output as "arduino or C++"
|
|
|
|
|
|
How to proceed :
|
|
|
|
- First, install the ai-trained-library in the arduino IDE
|
|
- Second, you need to push the "retrieve_data.ino" in you arduino board (The board we used is the Arduino Nano 33 BLE sense REV2)
|
|
- Then, in VSC (or other) run the code
|
|
|
|
In the end, the project works, but is open to a lot of improvment.
|
|
For exemple :
|
|
- The proximity captor dont have any influence on the stress
|
|
- The proximity should lower the base-stress, but it doesnt...
|
|
- The ai-recognition isn't always spot-on, leading to a variationof the stress when nothing is happening, this comes from the ai-training in edge-impulse directly
|
|
- There is a slight delay in reaction (but nothing too big)
|
|
- The sound level doesnt work because the mic is already used by the ai-recognition model
|