For audio and video transcripts, as an optimization, the matching algorithm could remove most of the transcript that is not within, say, ~ 1 minute of the specified time.  This would avoid false matches from earlier in the transcript and allow the the system to use less restrictive matches of bad translations.

Citing Text Example:

• a Wizard of Oz moment

 

But it Needs to match actual transcript:

• a wizard of oddz moment

 

Match Sensitivity

In this case, the system matches the quote properly, but one can imagine a situation where it does not match with the full context, but would match for a shorter excerpt of transcript with looser matching rules.

 

 

Transcript: 30 seconds of proximity

By narrowing down the transcript from the full transcript (saved), the sensitivity settings can be loosened, to match:

• oddz ~ Oz

also

1:26:58

confirmed what has been supposedly one of Donald Trump’s most outrageous uh you know lies or

1:27:06

exaggerations it’s fake news okay well if you are willing to be objective and

1:27:14

look at the evidence in this case and not give them a pass because our democracy is threatened you have to say

1:27:22

oh now I know not only that it’s fake but the exact methods by which they fake

1:27:27

it yeah yeah crazy it’s a wizard of oddz moment it really is yep you know the

1:27:34

curtain was pulled back absolutely so our story for this week uh

1:27:40

is a classic one of the last stories that Tolstoy ever wrote called after the ball written in 1903 and published

1:27:47

posthumously in 1911 and it’s a short but very

1:27:55

powerful uh beautiful dark story