Accessible Platform Architectures Working Group Teleconference

18 Jul 2018


janina, Joanmarie_Diggs, Becka11y, Léonie, JF, jasonjgw, MichaelC, gottfried, Mark_Hakkinen, Irfan_Ali, Peter, Becky


<scribe> scribe: Gottfried

Agenda Overview; Announcements

Janina: Welcome to the folks from ETS
... Topic is on how to get proper pronunciation for people who rely on speech.
... We will go over what they are proposing.
... How this would be in the stack and available to browsers.
... Then some other items (standard agenda)
... Any updates?
... I am confident that we have a place for this work in APA. Maybe in personalization tf. Work this activity in our new charter.

ETS Proposed Work

Markku: Director of digital accessibility of ETS.
... SSML has become an important topic for ETS.
... We deliver educational assessment / examinations.
... Text-to-speech pronunication is a critical requirement in the education classroom setting.
... Mispronounced words are problematic because there is a mismatch between was is being taught in the classroom vs. what is being rendered by TTS.
... We don't believe that using lexicons by TTS is the right solution.
... Rather the content matter experts / authors should apply the pronunciation rules.
... Vendors are often using hacking solutions, e.g. misuse of aria-label. Substitution of words in context.
... Often text expansions that make the synthesizer sound more correct.
... For 3 years we have trying to find a standards-based approach for this problem.
... What does it look like in HTML?
... We talked to AT and browser vendors and discussed ideas with them.
... The idea of inlining SSML in HTML was a no-starter.
... Vendors of read-aloud tools use data-* attributes.
... This seems to work.
... But this is not a standardized approach.
... We came up with our JSON-based model. Attribute aria-ssml with JSON content that would apply to the content of the HTML tag that the attribute is attached to.
... We think that this is the right approach for us.
... We would like to get serious feedback from W3C community.
... At the NFB convention, a blind student confirmed that pronunciation is very important for passing exams.

Peter: We have been working with pronunciation for 10 years - in proprietary ways.
... Tweak pronunciation for voice output. Database approach.
... data-ssml attribute

<Irfan> texthelp attribute example:

<Irfan> <span ssml="sub" alias="Every Student Succeeds Act">ESSA</span>

Peter: Hoster of the message responsible for putting the attribute on the tag.

Markku: Example from TextHelp uses multiple attributes.
... Drawback that you would sometimes need more than one or two attributes. Therefore, we propose JSON structure.

Leonie: We are trying to solve a problem by making authors responsible. Is this a good approach?
... The problem is caused by TTS.

Peter: Some prounciation issues are better handled by lexicons in synthesizers. But sometimes you have very specific authoring contexts, e.g. scientific terms, chemistry, mathematics.
... Also pausing between words.
... We got it wrong because we trusted the synthesizer to do it.
... Content authors know best how content should be spoken.

Leonie: Okay for specific content and audience.

Markku: Authoring platform with interface that allows content authors to highlight words that need pronunciation rules.
... Dialog allows them to pick the appropriate functions, without knowing JSON.

Janina: Lexicons could be improved, but they will not provide a complete solution. In particular, for geography and proper names.
... But we should not put all the burden on authors.
... Should address this more systematically.
... Character set switching on mobile apps.

<Irfan> +q

Janina: Testing should not rely on heuristics, and textbooks should not rely on heuristics either.
... Important for multi-language environments.

<Zakim> janina, you wanted to say lexicons won't be enough

Leonie: Example seems to be similar to <abbr> and @title. Seems straight forward.
... But how would you use this to correct pronunciation?

Markku: Substitution capability of SSML.
... Phonetic pronunciation also possible in SSML, e.g. using IPA.
... SSML can also control how numbers are spoken.
... Also pausing can be controlled, in milli-seconds.
... When supporting braille, many existing solutions go to sound and refreshable braille (does not make sense in braille).
... In SSML, you could have text displayed on braille that is different from the text that is sent to the synthesizer for pronunciation.

Joanie: With speech dispatchers, you can send SSML directly.
... Might be easy for AT to implement. I like this idea.

Markku: There was pushback on inlining SSML in HTML. But with JSON, it should be easier.

Leonie: Any response from other vendors other than TextHelp?

Markku: We have talked to a number of AT vendors, e.g. Freedom Scientific.
... Initial reaction: we are used to mispronunciations.
... When we talk about the use case of education, we get willingness to look into this.
... Also positive feedback from NVDA.
... James Craig (Apple) - JSON would allow to map speech rules to their internal markup.
... Talked to Microsoft and Google.
... Talked to Peter Korn (Amazon).
... Key is to get some kind of standards-based approach that they can use.

Leonie: Important to emphasize the educational aspect.
... With screenreaders, people are trying to come up with something that seems helpful, but it is not helpful.
... Terms in specific domains, e.g. music, are often pronounced completely different than regular.

Markku: Home automation and assistance products (e.g. Amazon Echo, Google Home) allow to use SSML to approve the user experience of spoken conversation.
... How can the average user benefit from this, but AT users not?

<Irfan> https://github.com/mhakkinen/SSMLinHTMLproposal

Janina: We agree that we have a problem that needs to be solved.
... Need to start the classic process for standardization.

Markku: The above URL describes the SSML approach that we are proposing.

Gottfried: In HTML, we have a lot of visual styling in three stages (author, UA, user). But that is missing today for speech styling.

Janina: Can ETS provide support from the wider publishing community?

Markku: Yes, we can provide resources, including in WEA(?)

Janina: Can we move that to personalization tf?

Joanie: Does not really fit. Personalization would be what an end user prefers. This is a more general approach for speech.

<Irfan> +1 to joanie ... ETS doesn't think personalization is a great fit

Joanie: It is not optional, not a choice, not a preference.
... Putting that in Personalization may not be an efficient way of moving it forward.

Janina: This should be available to everybody independent of AT.
... This is more mainstream than other things we are dealing with.

Michael: Don't see structural issues, only charter.
... Personalization and messaging are probably not good fits.
... This topic may be actually independent. Still thinking about this.
... Markku, you can assume that we can move forward.
... Encourage everybody to vote for the new APA charter.

Janina: Compelling thought that we have so little in speech styling compared to visual styling.
... Excited to see this happen.
... Charter approval should come soon, hopefully.

<scribe> ... New charter will allow for producing normative specs.

UNKNOWN_SPEAKER: Do we need time at TPAC to talk about this?

<Irfan> ETS will be at TPAC

Gottfried: Good to put this on the TPAC agenda since some people may join for mainstream use of this.

Janina: Need to bring this up to W3C.
... Thanks to ETS people who were present.

Charter Renewal Update

Janina: Yesterday, we had only a few votes.

Michael: 8 votes in support currently.

Janina: Will bring this up on the WAI coordination call.
... Everybody with voting rights, please vote.
... And remind others to vote.

TPAC 2018 https://www.w3.org/WAI/APA/wiki/Meetings/TPAC_2018

Janina: Conversation with people working on maths has been moved.

Joanie: Peter is not a fan of MathML. His community group is called "Getting math onto web pages"
... So it is about math in general, not MathML. But could include MathML.

Janina: RQTF continues discussion on supporting various knowledge domains.
... Create another AAM for defining other domains than math, e.g. music.
... Wide range of this kind of content.
... In RQTF, diagram work has been mentioned.
... My input: We need some kind of widget to support a good markup from domains that don't currently exist.
... Web platforms need to get into this discussion.
... Maybe there will be specific markup for every domain, e.g. chemistry, music.
... But my dream is to have a more generalized way.

Jason: RQTF is interested in this.
... Problem is more challenging because sometimes the specialized content is not discipline-specific.
... A11y APIs at UA level is very inflexible to adapt to changes that are specific to web apps or content domains.

<scribe> ... New technologies such as web components, accessibility object model are relevant here.

Janina: Hope we can discuss this at TPAC.
... In RQTF, research papers on this were mentioned.

CSS Update

new on TR http://www.w3.org/TR/tr-status-drafts.html

Michael: User timing and server timing
... We said no need to review

Janina closes the meeting.

Summary of Action Items

Summary of Resolutions

[End of minutes]

Minutes formatted by David Booth's scribe.perl version 1.152 (CVS log)
$Date: 2018/07/18 16:58:48 $

Scribe.perl diagnostic output

[Delete this section before finalizing the minutes.]
This is scribe.perl Revision: 1.152  of Date: 2017/02/06 11:04:15  
Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/

Guessing input format: Irssi_ISO8601_Log_Text_Format (score 1.00)

Succeeded: s/ahve/have/
Succeeded: s/maps/maths/
Default Present: janina, Joanmarie_Diggs, Becka11y, Léonie, JF, jasonjgw, MichaelC, gottfried, Mark_Hakkinen, Irfan_Ali, Peter, Becky
Present: janina Joanmarie_Diggs Becka11y Léonie JF jasonjgw MichaelC gottfried Mark_Hakkinen Irfan_Ali Peter Becky
Found Scribe: Gottfried
Inferring ScribeNick: gottfried
Found Date: 18 Jul 2018
People with action items: 

WARNING: IRC log location not specified!  (You can ignore this 
warning if you do not want the generated minutes to contain 
a link to the original IRC log.)

[End of scribe.perl diagnostic output]