First off, I see what you're going for and I like the idea! It actually reminds me of The Savage Gambit, a very old audio-only boxing title from the early 2000s which also included a trainer.
As for feedback:
I would personally move the offense keys to the right-hand side of the keyboard, that way you'd have more fluidity if you were going from offense to defense and back again. You could have the option to play with one hand available still as that is a useful accessibility consideration.
Furthermore, the idea of having "good timing" is currently nullified by the fact that I could press the button as soon as the prompt had given me enough information, with the punch being thrown part way through the "good timing" file playing.
With the ability to customise the speed of speech so the punches and moves you have to defend are spoken more quickly (by text to speech rather than what might be pre-recorded files), this would allow for those used to higher rates of speech to get more enjoyment out of this prototype as the game speed would comparatively increase in turn.
I definitely see potential here and like where this is going, well done.