Allow upload of transcriptions without timing information

Be able to import transcriptions without timing information and convert them into captions (words with timing information). ("Forced alignment")
See

https://github.com/pettarin/forced-alignment-tools/blob/master/README.md
https://medium.com/@hxianxian/forced-alignment-how-to-match-audio-with-a-transcript-via-machine-learning-dd19da8c0f04

We've had good success using CMU Sphinx in the past for 10 minute segments; however newer approaches may more accurate.
See
https://github.com/JoshData/cmusphinx-alignment-example
on how to get started with Sphinx.
Ideally this should work with 90 minute lectures content, which might include long silences and music, computer beeps etc.