How to implement text-to-speech matching?

Question

API

The task is to determine in the mobile application whether the person read the text from the screen of the device correctly.

Now I implemented it like this: the default Google speech recognizer built into my smartphone recognizes speech, and I compare the result already obtained with the text. In the long run, this is a bad option, because. sometimes this speech recognizer shamelessly lies.

There is also google speech api and amazon similar. They are smarter, they can send you different versions of the recognized text if in doubt, but they are expensive if you recognize a lot.

What other technologies are there that could be used to solve a similar problem? Perhaps other 3rd party apis or open source libraries? Or, perhaps, it is easier to write your own neural network, because the task of matching with text in theory should be much easier than just voice recognition?

Answer 1

Matching what is read with what is written?
Let one of them return the text of the speech to you, indicating the main points of the arrangement of the speech, and compare it at least by words. perfect if you recognize yourself.