How to calculate the probability value of a word?

[TNBT12g]

Hi everyone, I am new here, so I do not know how authors can calculate the value of probability. I guess they use one of the following methods:

1. Comparing the audio of a word in audio with a lot of sample sounds. Then, they count how many times it's correct and calculate the percentage. Example: With 100 sample sounds, if the word in the audio matches 50 sample sounds, the probability value is 50%.

2. Comparing the syllables of the word. Example: For the word 'contribute' - /kənˈtrɪb.juːt/. They check if the phoneme sounds are correct. If /kən/ sounds like /kan/ or /kon/, the word will only get 66.6% probability.

** If they are using the second approach, could you please explain how I can identify the incorrect phoneme?

I would like to receive results like:

• "birch": /bir/ may lack confidence, so its color is yellow, and /ch/ is the correct phoneme, so its color is green.
• "planks": /pl/ may lack confidence, so its color is green, and /anks/ is the correct phoneme, so its color is red. I came across this picture in a discussion, but it doesn't have a solution. Examples: "Agricultural," "depletion." I would greatly appreciate your assistance. Thank you so much.

I see the website name "apeuni" has this feature, you can see more example:

[jordimas]

The code that calculates this is here: https://github.com/guillaumekln/faster-whisper

Hopefully it helps