I've tried all of the closed captioning engines over the last few days. I think Cohere with Roll-up, 2 lines and a 2.0 sentence length is the best we've seen so far. I watched a show with my wife, being careful not to say "notice the subtitles," and she didn't have any complaints.
Ideally I'd like to run captioning after the show is complete, because the difference is noticeable. EDIT: this has already been solved here and I missed it! Synchronized Closed Captions for Channels
It doesn't look like there's a way to scan the file after it's finished and inject captions into the file's metadata the way ah4c is doing in real time. I tested a 30-minute show and captured an srt with Openwhisper-XXL and the large-v2 model in 8 minutes using the NVIDIA card on my Windows machine. The quality in terms of accuracy and appearing when the words are spoken instead of 2-3 seconds later is unmistakable. I didn't expect them to work, but neither an srt dropped next to the recording nor a muxed mp4+srt > mkv were recognized by Channels. I'm going to keep up not saying anything for a few days and see if the captions cause any complaints.
Thanks for your hard work, @anon38303228. Enjoy your rest!