The gaze/blink detection doesn't work so well (e.g. if your eye is still for some time, it "forgets" the gaze and becomes None). Find if it can be tuned, or find another library.

The system should ask the user to say a certain short phrase, randomly generated from a dictionary.

For this we need:

• Another Tkinter window to guide the task.
• A (short, for now) dictionary of phrases.
• Logic to handle the whole operation.

The window should also be timed, as described in #2

• Add Azure region as environment variable
• Flatten the configuration structure

Instead of having the "Shoot" button, the windows must have a visible timer (e.g. 3 seconds), after which:

• the frame is captured
• a sound is played (to help the user understand that the task is over)

Playsound says a thing that is unexpected:

playsound is relying on another python subprocess. Please use `pip install pygobject` if you want playsound to run more **efficiently.**

(also, the installation of pygobject failed on my system, so I cannot see the difference)

Find:

• why it happens
• if it is a problem
• how to solve it

Instead of just the plain task name, write something more.

Examples:

• "left" ==> "Look left"
• "happy" ==> "Do an happy face"

While a video operation is running (gaze, emotion), there should be a background blinking check.

In other words, there should be a certain ratio of blinks while the video window is open (configurable).

• Startup
• After speech recognition

Eg. "house" may be recognized as "how's"; "car" as "carl", etc.

Implement some edit distance measure to allow such errors.

The following config must be converted to seconds instead of frame ratios.

"blinking": {
        "min_threshold": 0.2,
        "max_threshold": 0.5
    },

Possible strategy:

• the callback function to Window must count the total number of analyzed frames
• you already know the window expiration time
• you calculate how many frame you have in 1 second
• you check whether the number of blinks is in the configured seconds range