Score_cutoff is inverted

[tjiho]

Hi !
Thanks for your work.

When using --correct-sentences , lower the CUTOFF is, more the output is open-ended.

If I set --correct-sentences 0.5 there will be no correction. Instead if I set --correct-sentences 200 every sentences will match my template.

From what I understood in the readme, the opposite is indicated:

When <CUTOFF> is large, speech recognition is effectively open-ended again.

[plashenkov]

If so, then that's correct and good. Let's say the difference between the recognized and reference phrases is 5. If you set a cutoff of, say, 10, then you'll consider it a match. Anything up to 10 will be a match. But if the difference between the phrases is greater than 10, then it's a mismatch, and the phrase will be passed on as it was recognized (unless you specified --limit-sentences).

Then the description in the README is incorrect, and it would be a good idea to correct it.

I read the README and was puzzled by the choice of such strange logic. I was about to open an issue, but then I found yours.