Think the idea of using a transcribing supplier is totally valid. Text to speech is clever, but not accurate. Guess I’d need to find one that’s happy with an NDA, but I expect that’s nothing extraordinary.
After this phase of the work however, I’m wondering if your suggested steps offer any real advantage over my current method. I think I’d certainly lose some flexibility it some subs needed amending in iterative drafts, and unsure whether it would speed things up much anyhow