Last week I shipped a YouTube section looper in Captrice, the guitar practice app I'm building. It's meant for learning songs by ear — you mark a section of the video to loop, play/pause, rewind as well as adjust playback speed. Besides being a looper, it's also a keyboard-driven controller for the YouTube video player embedded in the app so that you can operate it easily while playing guitar. Here's a short preview:

That's not what this post is about though. Building this functionality in Captrice made me wonder if it'd be similarly possible to control a YouTube video playing in a browser from inside emacs using keybindings. I have a simple use case — I use org-roam to take notes while learning. If the learning material is a YouTube video, I split the screen with the browser on one side and emacs on the other. The problem is that pausing or rewinding the video involves switching to the browser window, interacting with the player, and then switching back (and the same steps in reverse to resume). On MacOS the media controls on the touch bar come in handy for this use case. It's the one thing I've genuinely miss since moving to Linux.

Now if you know emacs, you can bet someone must have managed to get YouTube videos playing inside the editor. And sure enough, there are packages that turn emacs into a media player. But getting them to play YouTube videos requires additional dependencies (youtube-dl, mpv etc.) that I'd like to avoid if possible. I'm happy with the video playing in the browser as long as it can be controlled from emacs.

Turns out, it's quite trivial to achieve this. Most modern browsers on Linux implement the MPRIS2 D-Bus Interface out of the box. MPRIS2 stands for Media Player Remote Interfacing Specification 2.0. To put it simply, it's a standard Linux D-Bus interface through which a process can control an MPRIS-compatible media player running in another process.

When a web browser plays media, it registers itself as an MPRIS2 player on D-Bus and exposes its controls. The player can now be controlled from another process by sending messages over D-Bus. playerctl is a command line controller that makes this straightforward. For e.g. the following command will pause the video in the browser (if it's playing) and resume (if it's paused).

1

playerctl play-pause

Here's another handy command to rewind the video by 5 seconds

1

playerctl position 5-

There's however a bug (actually in YouTube's media player and not playerctl) that causes it to disregard the offset and always seek (ahead or behind depending on the sign) by 5s. This means the following command will also seek behind by 5s despite 15- being specified as the offset

1

playerctl position 15-

Thankfully, the position command accepts an absolute position (without the sign) too as input, so it's easy to work around this bug by doing the calculation ourselves.

1

playerctl position $(echo $(playerctl position) | awk '{v=int($1)-15; print (v<0)?0:v}')

playerctl position without any further argument returns the current position as a float. The awk command converts it to integer, subtracts 15 from it and clamps the result to 0 in case it's negative.

Coming back to emacs, all that's left to do now is to implement interactive functions that shell-out to playerctl and define keybindings for the same. We could also implement a MRPIS2 controller in elisp but it seems like an overkill.

My emacs config looks something like this:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12

(defun my/youtube-play-pause ()
  (interactive)
  (shell-command "playerctl play-pause"))

(defun my/youtube-rewind ()
  (interactive)
  (shell-command "playerctl pause && playerctl position $(echo $(playerctl position) | awk '{v=int($1)-15; print (v<0)?0:v}') && playerctl play"))

(define-prefix-command 'my/media-map)
(global-set-key (kbd "C-c m") 'my/media-map)
(define-key my/media-map (kbd "p") 'my/youtube-play-pause)
(define-key my/media-map (kbd "b") 'my/youtube-rewind)

Notice that the playerctl position command is wrapped in pause && … && play. I've observed that it works more reliably this way.

The youtube-rewind function can be parameterized to take the offset as input. But I have kept it simple for now and may add support for more control later if needed.

Here's a quick demo:

Due to the absence of audio, it's not immediately clear that the YouTube video on the left is indeed playing in the beginning. But you can follow the keybindings that appear in the right-bottom part of the screen as I type them. Also notice that emacs is always the active window for the entire duration which means I never switch to the browser window.

And thus, it's so much easier to take notes while learning from a video without interrupting the flow.