Subtitle files are arguably transformative and have a more benign character and purpose. They would have a significantly better fair use argument vs this. The purpose tends to either be to help the disabled or to assist non-native speakers in understanding the spoken dialogue track of the film. They also don’t directly compete with the underlying piece of copyrighted material (the screenplay) in the same market (publishing). A subtitle site could still easily end up on the losing end of a copyright suit but my point is this is far from an apples to apples comparison just because they both deal with text and audio/visual media.
Well OCRing books (for book search) was found to be fair use because “book search” and “books” are in different markets. So to the extent it is like book search that could be in its favor.
Your other example (text to speech and audio books) is significantly less transformative as audio books and books are basically the same or very related markets. (For example they are both sold in the same specialty stores)
