Improve Voice Generation

[Ketcap]

Is your feature request related to a problem? Please describe.
Right now voice is used to talk from the whole output, which is most of the time might not be meeded due to having some code, markdown or similiar in there.

Describe the solution you'd like
To clear the text before sending it to voice generation

Describe alternatives you've considered
N/A

Additional context
Voice generation solutions are expensive, and reading code does not makes sense while it's explaining the output.