[math-on-web] a11y TF meeting minutes, 2017-02-13

Hi everyone,

below are some minutes from the a11y task force meeting this week.

See you later for the regular call.
Peter.

# 2017-02-13 MathOnWeb a11y TF

* Dani: experimented with VO and SVGs
* Volker: experimented with SSML and speech synthesis API
* Dani: tested narrator for Edge, NVDA, ChromeVox
  * all slightly differently but seem consistent
  * had problems with VoiceOver
    * iOS
  * VoiceOver with SVG with role=math, then it will read the letters
    * with role=img, then alt-text is working
  * Peter: what does the SVG tech look like?
    * Dani: text in SVG + nested titles but nested titles do not work
  * Dani: two approaches:
    * a) text for entire formula
    * b) split so that you can explore
    * can't do both
    * Peter: I think that's intentional from WAI-ARIA
  * Volker: role=math not implemented well
    * Peter: spec says: all children are hidden, so nothing should be
exposed on a11y tree
* Peter: been working on deep lables (see codepen)
  * making progress
   * hacking around empty spans problem
   * some mixed version for SVG output
   * VO quite good but no navigation, nvda fails badly (reached out to
James)
   * Dani: is this customized per browsers?
     * Peter: no.
   * Dani: this is a standard MathJax rendering?
     * Peter: yes.
 * Volker: so does e.g., aria-label="two fifth" hide the content?
   * Peter: yes. Cf. James Teh's blog post on calculations
 * Peter: what I need to figure out is DOM exploration (i.e. what you can
see in dev tools)
    * VO is iffy with the demo
    * NVDA better in my experience but the demo fails
 * Volker: some argue in favor of clever linearization (a la MathPlayer)
   * Peter: nice but seems unrealistic to me
     * perhaps a theoretical new ARIA feature could help but unclear
   * Volker: it's a bit like levels of granularity (similar to sentence /
word / characters)
 * Peter: I will do a screencapture for VO, NVDA, also try VO+Chrome/FF
* Volker: been playing with speech synthesis
  * clearspeak in SRE by August => requires prosody changes (pausing)
  * similar to screenreaders using different prosody for e.g., capitals
  * speech synthesis being implemented in Edge
  * so instead of using live-regions we can push directly there
  * linux+FF fails (crashes pulseaudio)
  * http://caniuse.com/speech-synthesis, https://developer.mozilla.org/
en-US/docs/Web/API/Web_Speech_API
  * officially, speech API includes SSML, but does not seem implemented
  * initial test (https://github.com/zorkow/speech-rule-engine/blob/v2.0.
0-beta.1/tests/markup_test.js) works in watson (https://text-to-speech-demo.
mybluemix.net/)
  * if I create a live sample that should work, could you try it out in
your browsers?
    * => yes
* ACTIONS:
  * https://github.com/w3c/mathonwebpages/issues/12
  * https://github.com/w3c/mathonwebpages/issues/13
  * https://github.com/w3c/mathonwebpages/issues/11