4 ms·

I would approach this problem differently: tokenize the text and diff the stream of tokens (as opposed to a stream of characters).