Re: Help with Captioning

[Sacha Chua]

Another option is for you to send me what you've got and then take on a different talk instead. A few more have been added to the backstage area, and I remember the autogenerated captions for them were a lot cleaner. We want captioning to be fun for you too! :)

On Sun, Nov 6, 2022, 11:58 Sacha Chua <sacha@sachachua.com> wrote:

Hi Tom!

Hmm, yeah, the AI had a hard time with the realestate talk because of the run-on sentences. :) Would it help you to ignore splitting up the subtitles and just focus on correcting the text so that it has the right words? I can split up and re-time it afterwards.

On Sun, Nov 6, 2022, 11:47 Tom Purl <tom@tompurl.com> wrote:

Hi again. I just wanted to let you know that I'm still working on captioning the Real Estate talk. I haven't forgotten it, I'm just a bit slow at captioning I believe. But I'm getting used to it and really like using subed.el.

I wanted to get a baseline value from you. I find that I'm able to clean up about 5 minutes of captioning per hour. Is that especially slow? Granted, my entire talk is a demo, and so it's not a lot of full sentences or proper grammar. So maybe that's slowing me down too.

------- Original Message -------
On Wednesday, October 26th, 2022 at 7:17 PM, Tom Purl <tom@tompurl.com> wrote:

😀 sounds great, thank you for all of the help.

-------- Original Message --------
On Oct 26, 2022, 2:07 PM, Sacha Chua < sacha@sachachua.com> wrote:

All right, let's ease you in with editing a transcript, because that's just about editing. =)

I'll reassign the realestate talk from me to you. To work on it, you can download https://media.emacsconf.org/2022/backstage/emacsconf-2022-realestate--real-estate-and-org-table-formulas--daniel-gopar--normalized.webm and https://media.emacsconf.org/2022/backstage/emacsconf-2022-realestate--real-estate-and-org-table-formulas--daniel-gopar--main.vtt . You can open the VTT file in your favourite subtitle editor. If you don't have a favourite yet, you can do it in Emacs either as plain text or with the extra awesomeness of https://github.com/sachac/subed . (Disclaimer: I maintain it, so I'm interested in seeing if I can get it working for more people.)

Subed can synchronize with MPV, so you may want to install MPV as well. If you've loaded subed (and maybe subed-vtt, if I've messed up my autoloads and it doesn't load automatically), you should be able to open the file and then use C-c C-v (subed-mpv-find-video) to play the video. You can use M-SPC to toggle pausing and M-j to jump to the nearest subtitle. You can move your point to a word and then use M-. to split the subtitle at the current playback position or at a point proportional to where your point is in the subtitle if the current subtitle isn't the one playing. M-m merges the subtitle with the next one, and M-M merges it with the previous one. You can see that I've shortened the subtitles until about 6:55, changed things like "Control c Control c" to C-c C-c, turned things into sentences, etc.

If you want to get even fancier when it comes to splitting, you can download https://media.emacsconf.org/2022/backstage/emacsconf-2022-realestate--real-estate-and-org-table-formulas--daniel-gopar--main.srv2 and use the newly-added word data support from subed-word-data.el (in the same repo, version 1.0.16) with M-x subed-word-data-load-from-file to split the subtitles. It should use a greenish text when the word data has been successfully lined up with the caption. When you split with the usual M-. (subed-split-subtitle), it'll use the word timing data if available. The code is pretty new, so let me know if you run into problems.

Feel free to e-mail me partial edits if you want to check in or ask questions - you don't have to wait until you finish the whole file. =) Let's get you warmed up!

Sacha Chua - sacha@sachachua.com - Blog: sachachua.com

On Wed, Oct 26, 2022 at 2:48 PM Tom Purl <tom@tompurl.com> wrote:

Thank you for your quick response. I'm happy to help with a project that I enjoy so much.

I am also very​ new to transcription software but it seems like fun. I'm a DevOps engineer for a living and have written a half-dozen Ansible playbooks before. I also spend about half my time writing Python scripts in Emacs.

I'll leave prioritization up to you folks. I'd probably be the most effective with the Ansible provisioning stuff, but any technical task sounds fun 🙂 In the meantime I'll start looking at the "backstage" area.

------- Original Message -------
On Wednesday, October 26th, 2022 at 4:15 PM, Sacha Chua <sacha@sachachua.com> wrote:

That's fantastic! I'll send you an e-mail with the information for the backstage area so you can have a look around, and I look forward to learning from your experience. We'll definitely appreciate all the captioning help we can get once the talks start coming in!

I'm new to using whisper, lhotse, and torchaudio, so there are probably all sorts of improvements we can make to https://git.emacsconf.org/emacsconf-ansible/tree/roles/caption/templates/process-captions.py . For example, I'm curious about how we can get it to do common edits for us, like splitting subtitles at pauses. I'm also working on adding more stuff to subed.el to take advantage of word-level data. Let me know if that sort of scripting floats your boat!

Also on the scripting/automation side, we're experimenting with using Ansible this year so that we can make the configuration easier to reproduce. Is that also something you like or want to play around with?

Sacha

On Wed, Oct 26, 2022, 10:47 Tom Purl <tom@tompurl.com> wrote:

I believe I can help with captioning. I've used whisper a bit and have quite a bit of scripting and automation experience.

Thank you,

Tom Purl