|
|
* Things to check or decide
- Can we stream from multiple rooms without getting thoroughly confused?
- Assign rooms shortly before presentation
- *Shortly before presentation:* accommodates reshuffling
- Generating static images
- Before the presentation: Talk title, speaker
- Speaker checked in
- After the presentation: pointer to main conference stream
- After the conference: conference has ended, we'll take some time to prepare the recording
* Workflows
O1 - main organizer, O2 - secondary organizer or volunteer
*** Overview
- Check in via IRC at #emacsconf-org
*** Before the conference
- Do tech checks and get alternative ways to contact speakers (phone number? IRC nick? Something that goes ding?)
- Install Mute Tab extension if desired
*** Start of streaming
- O1 starts streaming
- O2 starts low-resolution mirror (see infra.org in private wiki)
*** Tech check
- Explain process
- Test audio, webcam, screensharing, collaborative pad
- Check if comfortable checking into IRC
- Ask about Q&A preference; OR:
- live Q&A
- Q&A over pad or IRC
- no Q&A
- Get IRC nick and emergency contact information
- Record name pronunciation
*** Shortly before the presentation
- Speaker checks in via IRC
- O2 directs speaker to available room
- Speaker joins talk room
- O2 makes speaker presenter and moderator, does tech check
- Speaker tries screen sharing and webcam
- O2 briefs speaker on process, including having the speaker say their name clearly and do a quick intro to their talk
- O2 notifies O1 with link to the talk room
- O1 joins meeting and gives go-ahead
- O2 starts recording in BBB
- O2 announces on IRC
*** During the presentation
- OR:
- Live presentation?
- Speaker presents, keeping an eye on the collaborative pad for questions
- O1 stays with speaker to stream and to help with questions and timing
- Prerecorded? O1 plays prerecorded video on computer
- If there is another speaker, O2 moves to next room for setup
*** Q&A
- O1 joins speaker's room if not already there
- O1 gives go-ahead to start Q&A
- Speaker reads questions off the pad or gets questions from O1
- O1 goes to next talk when ready
*** Scenarios
**** Prerecorded presentations
- Amin will play it on his computer and stream from there
**** Tech issues
- If can't be easily resolved, play pre-recorded talk early and try again later (or follow up)
- Stream a technical issues slide to the end point
* Splitting up the stream recording into individual files
** Timing
** Day 1
| 00 | Opening remarks | 13:04 | 20:08 |
| 02 | An Emacs Developer Story: From User to Package Maintainer | 27:24 | 51:39 |
| 03 | Q&A Idea to Novel Superstructure: Emacs for Writing | 1:00:10 | 1:09:21 |
| 05 | Q&A Bard Bivou(m)acs - Building a bandcamp-like page for an album of music | 1:41:30 | 1:55:12 |
| 06 | Trivial Emacs Kits | 1:57:25 | 2:11:04 |
| 07 | Q&A Beyond Vim and Emacs: A Scalable UI Paradigm | 2:32:02 | 2:36:34 |
| 08 | Building reproducible Emacs | 2:38:03 | 2:55:21 |
| 09 | Q&A Orgmode - your life in plain text | 4:22:00 | 4:27:48 |
| 12 | One Big-ass Org File or multiple tiny ones? Finally, the End of the debate! | 4:55:23 | 5:11:38 |
| 13 | Experience Report: Steps to "Emacs Hyper Notebooks" | 5:13:07 | 5:25:11 |
| 15 | Moving from Jekyll to OrgMode, an experience report | 5:47:45 | 6:04:16 |
| 16 | Org-roam: Presentation, Demonstration, and What's on the Horizon | 6:06:05 | 6:27:55 |
| 17 | Org-mode and Org-Roam for Scholars and Researchers | 6:29:00 | 6:50:14 |
| 18 | Org-roam: Technical Presentation | 6:51:20 | 7:21:45 |
| 40 | Closing remarks part 1 | 7:24:01 | 7:27:23 |
| 20 | OMG Macros | 7:28:00 | 7:50:36 |
| 40 | Closing remarks part 2 | 7:51:48 | 8:04:25 |
** Commands used for splitting
#+begin_src sh :eval no
ffmpeg -y -i main.webm-2020-11-28--08-48.webm -ss 13:04 -to 20:08 -c:a copy -c:v copy emacsconf-2020--00-opening-remarks.webm
ffmpeg -y -i main.webm-2020-11-28--08-48.webm -ss 27:24 -to 51:39 -c:a copy -c:v copy emacsconf-2020--03-an-emacs-developer-story-from-user-to-package-maintainer--leo-vivier.webm
ffmpeg -y -i main.webm-2020-11-28--08-48.webm -ss 1:00:08 -to 1:09:22 -c:a copy -c:v copy emacsconf-2020--03-idea-to-novel-superstructure-emacs-for-writing--questions--bala-ramadurai.webm
ffmpeg -y -i main.webm-2020-11-28--08-48.webm -ss 1:41:28 -to 1:55:13 -c:a copy -c:v copy emacsconf-2020--05-bard-bivoumacs-building-a-bandcamp-like-page-for-an-album-of-music--questions--grant-shangreaux.webm
ffmpeg -y -i main.webm-2020-11-28--08-48.webm -ss 1:57:24 -to 2:11:05 -c:a copy -c:v copy emacsconf-2020--06-trivial-emacs-kits--corwin-brust.webm
ffmpeg -y -i main.webm-2020-11-28--08-48.webm -ss 2:32:00 -to 2:36:35 -c:a copy -c:v copy emacsconf-2020--07-beyond-vim-and-emacs-a-scalable-ui-paradigm--questions--sid-kasivajhula.webm
ffmpeg -y -i main.webm-2020-11-28--08-48.webm -ss 4:55:00 -to 5:11:38 -c:a copy -c:v copy emacsconf-2020--12-one-big-ass-org-file-or-multiple-tiny-ones-finally-the-end-of-the-debate--leo-vivier.webm
ffmpeg -y -i main.webm-2020-11-28--08-48.webm -ss 5:47:44 -to 6:04:17 -c:a copy -c:v copy emacsconf-2020--15-moving-from-jekyll-to-orgmode-an-experience-report--adolfo-villafiorita.webm
ffmpeg -y -i main.webm-2020-11-28--08-48.webm -ss 6:06:00 -to 6:27:56 -c:a copy -c:v copy emacsconf-2020--16-org-roam-presentation-demonstration-and-whats-on-the-horizon--leo-vivier.webm
ffmpeg -y -i main.webm-2020-11-28--08-48.webm -ss 7:24:00 -to 7:27:09.50 -c:a copy -c:v copy emacsconf-2020--40-closing-remarks-part-1.webm
ffmpeg -y -i main.webm-2020-11-28--08-48.webm -ss 7:28:00 -to 7:50:50 -c:a copy -c:v copy emacsconf-2020--20-omg-macros--corwin-brust.webm
ffmpeg -y -i main.webm-2020-11-29--08-44.webm -ss 15:40 -to 27:27 -c:a copy -c:v copy emacsconf-2020--41-opening-remarks.webm
ffmpeg -y -i main.webm-2020-11-29--08-44.webm -ss 34:48 -to 1:03:54.75 -c:a copy -c:v copy emacsconf-2020--22-powering-up-special-blocks--musa-al-hassy.webm
ffmpeg -y -i main.webm-2020-11-29--08-44.webm -ss 1:30:40 -to 1:49:18 -c:a copy -c:v copy emacsconf-2020--23-incremental-parsing-with-emacs-tree-sitter--questions--tuan-anh-nguyen.webm
ffmpeg -y -i main.webm-2020-11-29--08-44.webm -ss 4:33:04 -to 4:37:30 -c:a copy -c:v copy emacsconf-2020--26-emacs-as-a-highschooler-how-it-changed-my-life--questions--pierce-wang.webm
ffmpeg -y -i main.webm-2020-11-29--08-44.webm -ss 4:50:00 -to 5:59:30 -c:a copy -c:v copy emacsconf-2020--28-welcome-to-the-dungeon--erik-elmshauser-corwin-brust.webm
ffmpeg -y -i main.webm-2020-11-29--08-44.webm -ss 6:08:20 -to 6:12:42.40 -c:a copy -c:v copy emacsconf-2020--30-a-tour-of-vterm--questions--gabriele-bozzola-sbozzolo.webm
ffmpeg -y -i main.webm-2020-11-29--08-44.webm -ss 6:25:18 -to 6:31:04.90 -c:a copy -c:v copy emacsconf-2020--31-lakota-language-and-emacs--questions--grant-shangreaux.webm
ffmpeg -y -i main.webm-2020-11-29--08-44.webm -ss 6:58:20 -to 7:20:38 -c:a copy -c:v copy emacsconf-2020--33-maxima-a-computer-algebra-system-in-emacs--fermin.webm
ffmpeg -y -i main.webm-2020-11-29--08-44.webm -ss 6:58:20 -to 7:20:38 -c:a copy -c:v copy emacsconf-2020--33-maxima-a-computer-algebra-system-in-emacs--fermin.webm
ffmpeg -y -i main.webm-2020-11-29--08-44.webm -ss 7:55:44 -to 8:02:02 -c:a copy -c:v copy emacsconf-2020--35-waveing-at-repetitive-repetitive-repetitive-music-zmusic--questions--zachary-kanfer.webm
ffmpeg -y -i main.webm-2020-11-29--08-44.webm -ss 8:03:32 -to 8:40:01.10 -c:a copy -c:v copy emacsconf-2020--42-closing-remarks.webm
#+end_src
Run this code before a copy of the ffmpeg scripts (adjusting the value of adjust as needed) and it will show only the lines that need tweaking.
#+begin_src emacs-lisp :eval no
(save-excursion
(while (re-search-forward "^ffmpeg.*?-ss +\\([^ ]++?\\) +.*$" nil t)
(let* ((adjust 4)
(from (save-match-data
(let ((s (match-string 1)) num)
(if (string-match "\\(\\([0-9]+\\):\\)?\\([0-9]+\\):\\([0-9]+\\)\\(.[0-9]+\\)?" s)
(progn
(setq num
(+ (* 3600 (string-to-number (or (match-string 2 s) "0")))
(* 60 (string-to-number (or (match-string 3 s) "0")))
(string-to-number (or (match-string 4 s) "0"))
(string-to-number (concat "0" (or (match-string 5 s) "")))))
(if (> (% num adjust) 0)
(number-to-string (- num (% num adjust)))
nil))
s)))))
(if from
(replace-match from nil nil nil 1)
(replace-match "")))))
#+end_src
ffmpeg -y -i main.webm-2020-11-29--08-44.webm -ss 23116 -to 6:31:04.90 -c:a copy -c:v copy emacsconf-2020--31-lakota-language-and-emacs--questions--grant-shangreaux.webm
Thanks to SirVolta and bandali for figuring out keyframe issue!
*** Additional notes
I've decided to use a etherpad document for this to make it possible for everyone to contribute and fix mistakes. Hope that's ok for everybody interested.
And hopefully I won't destroy too much by accidentally hitting Emacs keys :')
Feel free to edit at will. The goal of this document is to learn as much as possible about video manipulation
using ffmpeg.
**** video compression: keyframes and deltaframes
https://blog.video.ibm.com/streaming-video-tips/keyframes-interframe-video-compression/#keyframe
The problem is actually really simple. Most video contains a lot of non changing background images. It would be wasteful to re-encode all that every single frame. This is solved by only sometimes encoding the entire frame (keyframe, also called interframe or i-frame) and use the difference between the video in that frame and the current frame most of the time (deltaframe). The exact ratio between keyframes and deltaframes is a configuration option of the encoder. But this means that if you grab just a single frame, chances are high that it is a deltaframe and you would need at least one more keyframe or, depending on the codec, even all of the frames before, after or in between to restore the original frame content. This is also why going frame by frame backwards in MPV is so slow, and VLC doesn't even support it. It has to go all the way backwards to the last keyframe and reconstruct from there.
This means that when copying data, you can only start and stop on keyframes as deltaframes need the data before and possibly after it to work.
The VP8 codec used appears to only look backwards to a previous keyframe.
This explains why all endings were perfect.
To be reliable with all codecs, I consider the data after keyframes as important as the data before.
**** Possible solution
I'm thinking about splitting the rendering process in three pieces:
1. cut the time before the first keyframe and after the last keyframe using re-encoding
2. from that point (so between the first and last keyframe) cut using copy codec
3. merge the three
This way there is the least possible amount of re-encoding done while still having frame perfect timing.
It would mean figuring out a way to coax ffmpeg into splitting exactly on iframes and knowing where this is.
I think it's okay to cut to the keyframe boundary before the start timestamp. Up to three extra seconds of content never hurt anyone. =)
I'm currently testing a proof-of-concept script to do exactly this.
Command to extract all keyframe times:
AV_LOG_FORCE_NOCOLOR=1 ffmpeg -i input -vf select='eq(pict_type\,PICT_TYPE_I)' -f null NUL -loglevel debug 2>&1 | grep -F 'select_out:0' | awk '{print $4 " " $6}'
It should be a lot more efficient (or more like less inefficient) using start and end times.
Using this data, it should be possible to glue everything together in a script. Although it is going to be annoying dealing with time in seconds and HH:MM:SS.ms format. time<->string conversions. Yaay...
Or just go with seconds everywhere. Human readable formats are nice but kinda pointless now you can't edit them without causing keyframe issues.
**** resources
(all links work with LibreJS enabled)
https://blog.video.ibm.com/streaming-video-tips/keyframes-interframe-video-compression/
https://blog.streamspot.com/blog/compression-codecs-keyframes-and-the-basics-of-stream-quality
https://blog.superuser.com/2012/02/24/ffmpeg-the-ultimate-video-and-audio-manipulation-tool/
https://superuser.com/questions/138331/using-ffmpeg-to-cut-up-video
http://blog.webmproject.org/2010/05/inside-webm-technology-vp8-alternate.html
|
