Math Reading Issues (pronunciation of x^3, y^3)


Joy Holly
 

I am building an online Algebra course for a client that is inside the Canvas LMS. We are currently having an accessibility review and NVDA (which with MathPlayer installed reads math better, overall, than JAWS) is having trouble reading cubed variables. For example, "x^3" reads as "skubt" and "y^3" reads as "eekup". (But "x^2" will read "x squared" as expected, and "x^9" will read "x to the ninth" as expected. JAWS can read it, so the coding isn't wrong, but this is throwing an error in our accessibility review. I am aware individual users can change the voice or tweak the reading settings in NVDA.  But this is a public education course so we need to present the content to everyone in the best initial state possible. So I am wondering if this misreading of x and y variables is a common bug with NVDA, common enough that most NVDA users will be familiar with these reading anomalies? The testing is being done with NVDA/MathPlayer, in Chrome. Would appreciate any feedback about known NVDA math reading anomalies.

Thank you!

Joy Holly


Rui Fontes
 

First things we need to know:

1 - What language?

2 - What synth?


Rui Fontes


Às 16:35 de 19/08/2022, sjoyholly@... escreveu:

I am building an online Algebra course for a client that is inside the Canvas LMS. We are currently having an accessibility review and NVDA (which with MathPlayer installed reads math better, overall, than JAWS) is having trouble reading cubed variables. For example, "x^3" reads as "skubt" and "y^3" reads as "eekup". (But "x^2" will read "x squared" as expected, and "x^9" will read "x to the ninth" as expected. JAWS can read it, so the coding isn't wrong, but this is throwing an error in our accessibility review. I am aware individual users can change the voice or tweak the reading settings in NVDA.  But this is a public education course so we need to present the content to everyone in the best initial state possible. So I am wondering if this misreading of x and y variables is a common bug with NVDA, common enough that most NVDA users will be familiar with these reading anomalies? The testing is being done with NVDA/MathPlayer, in Chrome. Would appreciate any feedback about known NVDA math reading anomalies.

Thank you!

Joy Holly


 

I'd add to Rui's queries, "Have you tried different synths to see if the behavior carries over?"

This is very often "a synth thing."
--

Brian - Windows 10, 64-Bit, Version 21H2, Build 19044  

The difference between a top-flight creative man and the hack is his ability to express powerful meanings indirectly.

         ~ Vance Packard


Joy Holly
 

NVDA (with MathPlayer) is reading MathML placed within the html on course pages inside the Canvas LMS. Canvas also has the Mathjax engine implemented. No idea about synth. It is most recent version of NVDA, with MathPlayer also installed. No customizations. "Out of the box." Testing is in Chrome. Expected " x cubed" reading mashes pronunciation of the variable with the pronunciation of "cubed", so reads "x cubed" as "skubrrt" for example. Recognizes it as math but is botching the reading. For comparison, "x squared" reads fine, as does "x to the ninth". 

Is this a known difficulty, to the point that most NVDA users who are reading math routinely might be familiar with the issue? (And have an idea how to tweak NVDA for better results?) Or is there a certain way the MathML presentation must be in order to make the variable and wxponent reading more explicit? (Kind of like the use of invisible times, in , MathML, which makes explicit that variables adjacent to numbers are multiplied?) 

I am trying to get a feeling for if there are action points we need to take on the developer side, or if NVDA, like JAWS, just has quirks that users learn how to work around and this pronunciation issue with math is a wll known one, or...?

On Fri, Aug 19, 2022, 12:02 PM Rui Fontes <rui.fontes@...> wrote:

First things we need to know:

1 - What language?

2 - What synth?


Rui Fontes


Às 16:35 de 19/08/2022, sjoyholly@... escreveu:
I am building an online Algebra course for a client that is inside the Canvas LMS. We are currently having an accessibility review and NVDA (which with MathPlayer installed reads math better, overall, than JAWS) is having trouble reading cubed variables. For example, "x^3" reads as "skubt" and "y^3" reads as "eekup". (But "x^2" will read "x squared" as expected, and "x^9" will read "x to the ninth" as expected. JAWS can read it, so the coding isn't wrong, but this is throwing an error in our accessibility review. I am aware individual users can change the voice or tweak the reading settings in NVDA.  But this is a public education course so we need to present the content to everyone in the best initial state possible. So I am wondering if this misreading of x and y variables is a common bug with NVDA, common enough that most NVDA users will be familiar with these reading anomalies? The testing is being done with NVDA/MathPlayer, in Chrome. Would appreciate any feedback about known NVDA math reading anomalies.

Thank you!

Joy Holly


 

Hi,

Another datapoint to consider: have you tried with Firefox?

Cheers,

Joseph


Rui Fontes
 

You did not answer in what language NVDA is working...

English? Don't seems so...


Rui Fontes


Às 18:16 de 19/08/2022, Joy Holly escreveu:

NVDA (with MathPlayer) is reading MathML placed within the html on course pages inside the Canvas LMS. Canvas also has the Mathjax engine implemented. No idea about synth. It is most recent version of NVDA, with MathPlayer also installed. No customizations. "Out of the box." Testing is in Chrome. Expected " x cubed" reading mashes pronunciation of the variable with the pronunciation of "cubed", so reads "x cubed" as "skubrrt" for example. Recognizes it as math but is botching the reading. For comparison, "x squared" reads fine, as does "x to the ninth". 

Is this a known difficulty, to the point that most NVDA users who are reading math routinely might be familiar with the issue? (And have an idea how to tweak NVDA for better results?) Or is there a certain way the MathML presentation must be in order to make the variable and wxponent reading more explicit? (Kind of like the use of invisible times, in , MathML, which makes explicit that variables adjacent to numbers are multiplied?) 

I am trying to get a feeling for if there are action points we need to take on the developer side, or if NVDA, like JAWS, just has quirks that users learn how to work around and this pronunciation issue with math is a wll known one, or...?

On Fri, Aug 19, 2022, 12:02 PM Rui Fontes <rui.fontes@...> wrote:

First things we need to know:

1 - What language?

2 - What synth?


Rui Fontes


Às 16:35 de 19/08/2022, sjoyholly@... escreveu:
I am building an online Algebra course for a client that is inside the Canvas LMS. We are currently having an accessibility review and NVDA (which with MathPlayer installed reads math better, overall, than JAWS) is having trouble reading cubed variables. For example, "x^3" reads as "skubt" and "y^3" reads as "eekup". (But "x^2" will read "x squared" as expected, and "x^9" will read "x to the ninth" as expected. JAWS can read it, so the coding isn't wrong, but this is throwing an error in our accessibility review. I am aware individual users can change the voice or tweak the reading settings in NVDA.  But this is a public education course so we need to present the content to everyone in the best initial state possible. So I am wondering if this misreading of x and y variables is a common bug with NVDA, common enough that most NVDA users will be familiar with these reading anomalies? The testing is being done with NVDA/MathPlayer, in Chrome. Would appreciate any feedback about known NVDA math reading anomalies.

Thank you!

Joy Holly


 

On Fri, Aug 19, 2022 at 01:24 PM, Rui Fontes wrote:

You did not answer in what language NVDA is working...

English? Don't seems so...

 

-
I have no idea what makes you say that.  All indications are that English is the language in use and with very little reading between the lines.  The query being posted here, in English, with the expected (and sometimes obtained) announcements from NVDA documented in English cannot be considered coincidental.

A confirmation won't hurt, however.
--

Brian - Windows 10, 64-Bit, Version 21H2, Build 19044  

The difference between a top-flight creative man and the hack is his ability to express powerful meanings indirectly.

         ~ Vance Packard


 

On Fri, Aug 19, 2022 at 01:17 PM, Joy Holly wrote:
I am trying to get a feeling for if there are action points we need to take on the developer side, or if NVDA, like JAWS, just has quirks that users learn how to work around and this pronunciation issue with math is a well known one, or...?
-
And everyone here is trying to help you with narrowing down what, precisely, is at play here.

For any screen reader, not just NVDA, it is the synthesizer that's in use that decides precisely how whatever is passed to it is pronounced.  There are times a specific synthesizer just doesn't pronounce certain things at all well or as expected, and the first thing to do both for diagnosis and possible remediation is to try the same things you've been doing but with a different synthesizer in use.  If another synth says exactly what you'd expect, then you have clear evidence it's "a synth thing."  That's not something that developers would normally control, but the end-user does.

It could be, as has been suggested, that Chrome is not playing well with MathML under a very specific circumstance, so trying exactly what you've been doing using Firefox instead is of direct diagnostic value.

You've asked an excellent question, but there is no easy "one size fits all" answer.  The diagnostic process of elimination must take place.
--

Brian - Windows 10, 64-Bit, Version 21H2, Build 19044  

The difference between a top-flight creative man and the hack is his ability to express powerful meanings indirectly.

         ~ Vance Packard


Rui Fontes
 

Hello!


To you "eekup" seems an english message?

For me it is not an english message...


Rui Fontes



Às 18:33 de 19/08/2022, Brian Vogel escreveu:

On Fri, Aug 19, 2022 at 01:24 PM, Rui Fontes wrote:

You did not answer in what language NVDA is working...

English? Don't seems so...

 

-
I have no idea what makes you say that.  All indications are that English is the language in use and with very little reading between the lines.  The query being posted here, in English, with the expected (and sometimes obtained) announcements from NVDA documented in English cannot be considered coincidental.

A confirmation won't hurt, however.
--

Brian - Windows 10, 64-Bit, Version 21H2, Build 19044  

The difference between a top-flight creative man and the hack is his ability to express powerful meanings indirectly.

         ~ Vance Packard


 

On Fri, Aug 19, 2022 at 01:42 PM, Rui Fontes wrote:

To you "eekup" seems an english message?

For me it is not an english message...

-
Which is precisely the point and the problem.  It's isolated to cubed variables, and one of them, the X one, is at least far closer to what it should be than the Y one is.

All other instances referenced make it crystal clear that the announcement is both correct, and in English.

I don't know how that could possibly have been clearer from the initial problem statement, message  .  This is one of the best-written complex problem statements I've ever had the privilege to read on this group.
--

Brian - Windows 10, 64-Bit, Version 21H2, Build 19044  

The difference between a top-flight creative man and the hack is his ability to express powerful meanings indirectly.

         ~ Vance Packard


 

On Fri, Aug 19, 2022 at 11:58 AM, Joy Holly wrote:
I am aware individual users can change the voice or tweak the reading settings in NVDA.  But this is a public education course so we need to present the content to everyone in the best initial state possible.
-
Joy,

And while I understand this need, entirely, the reality is that you will not have control over what voice or voices some random end user has chosen to use with their respective screen readers.  It will be critical to note that an issue such as you've described can occur with certain synthesizers, and that another should be tried.  Offering specific examples of synths where everything is known to work correctly should be a part of the documentation that comes with the course.

There are certain things, unless you are setting up the entire environment from scratch yourself, that you will not have control over.  Documenting "known issues" such as this in a "Before You Start" type cheat sheet goes a long way toward warding off questions that could be avoided.
--

Brian - Windows 10, 64-Bit, Version 21H2, Build 19044  

The difference between a top-flight creative man and the hack is his ability to express powerful meanings indirectly.

         ~ Vance Packard


Joy Holly
 

Want to preface by saying in no way was I hoping anyone here would do troubleshooting I can do, was just looking for feedback to the question "is this sort of math reading error common/known" so I could rule out it being something with the LMS, with the version of Mathjax used, etc. to widen my knowledgebase before knowing what to try out next.

I will now answer my own question, being given the hint by you all to experiment with the set synths and voices. I had not done this because the goal on our side is to provide accessible and readable math, with the responsibility for the customization needed for this screenreader or that being with the user.

This reading error occurs also in FF, and I should have said this (but Chrome is our target browser since it is the one used by the schools.)

The default synth setting is what we are using since we adjusted nothing after installation. It is Windows OneVoice with the voice Microsoft David. I tries the other two OneVoice voice options; the pronunciation issue remained. So then I changed the synth setting to Microsoft Speech API v. 5, using the Microsoft Dave voice, and now it reads "x cubed" as expected! 

Thank you for the breadcrumbs leading me to the information I needed most, which was essentially confirmation that the math is accessible, and the reading issue here lies with NVDA specifications that need tweaking.

Very grateful!

Joy


On Fri, Aug 19, 2022, 1:50 PM Brian Vogel <britechguy@...> wrote:
On Fri, Aug 19, 2022 at 01:42 PM, Rui Fontes wrote:

To you "eekup" seems an english message?

For me it is not an english message...

-
Which is precisely the point and the problem.  It's isolated to cubed variables, and one of them, the X one, is at least far closer to what it should be than the Y one is.

All other instances referenced make it crystal clear that the announcement is both correct, and in English.

I don't know how that could possibly have been clearer from the initial problem statement, message  #98716.  This is one of the best-written complex problem statements I've ever had the privilege to read on this group.
--

Brian - Windows 10, 64-Bit, Version 21H2, Build 19044  

The difference between a top-flight creative man and the hack is his ability to express powerful meanings indirectly.

         ~ Vance Packard


Joy Holly
 

Yes, I will certainly make this recommendation as I 100% agree with you that this should be provided to user (especially for the NVDA screenreader, since it is free and the one most likely to be used in a public school setting). However, early documentation I included in a user info doc was edited down to generalities by the powers that be (I'm assuming for fear of having too much specificity that would then need editing oversight as these specifications change over time) so I may lose this battle, too. I'm mostly, at this stage, trying to help the accessibility engineers be able to identify what he fixable issues are because I had a situation where the reading of the exponents was on my fix list, and uh, the code is right, so there isn't any action point there by me, on the development end. Information we need to provide to the user is another very important conversation.

Thank you for your feedback!
-Joy

-Joy


On Fri, Aug 19, 2022 at 2:45 PM Brian Vogel <britechguy@...> wrote:
On Fri, Aug 19, 2022 at 11:58 AM, Joy Holly wrote:
I am aware individual users can change the voice or tweak the reading settings in NVDA.  But this is a public education course so we need to present the content to everyone in the best initial state possible.
-
Joy,

And while I understand this need, entirely, the reality is that you will not have control over what voice or voices some random end user has chosen to use with their respective screen readers.  It will be critical to note that an issue such as you've described can occur with certain synthesizers, and that another should be tried.  Offering specific examples of synths where everything is known to work correctly should be a part of the documentation that comes with the course.

There are certain things, unless you are setting up the entire environment from scratch yourself, that you will not have control over.  Documenting "known issues" such as this in a "Before You Start" type cheat sheet goes a long way toward warding off questions that could be avoided.
--

Brian - Windows 10, 64-Bit, Version 21H2, Build 19044  

The difference between a top-flight creative man and the hack is his ability to express powerful meanings indirectly.

         ~ Vance Packard


 

On Fri, Aug 19, 2022 at 02:49 PM, Joy Holly wrote:
Thank you for the breadcrumbs leading me to the information I needed most, which was essentially confirmation that the math is accessible, and the reading issue here lies with NVDA specifications that need tweaking.
 
-
Joy,

I am not trying to beat a dead horse here, as I've been where you are myself.  You can say, accurately, that an NVDA setting for voice may need tweaking, but it is critical to note that the issue does not lie with NVDA (nor JAWS, nor Narrator) but with the synthesizer in use.

There is nothing for the developers of NVDA (or JAWS, or Narrator) to fix in an instance like this.  A report could be made to the developers of the wonky synthesizer so they are aware of this odd bobble when reading mathematical expressions, but it would be up to the synth developers to fix it, not the developers of the screen readers that can send text to that synth.

This is not as hair-splitting as it seems, as getting issues under the noses of the right people to fix them is the quickest way to get them fixed (and that's only if the brains above those noses care). 
--

Brian - Windows 10, 64-Bit, Version 21H2, Build 19044  

The difference between a top-flight creative man and the hack is his ability to express powerful meanings indirectly.

         ~ Vance Packard


Joy Holly
 

I would have no idea how to make a report to the developers of the Microsoft OneCore voices synthesizer, although I agree it would be a helpful and noble thing to do. Is there a Wonky Synthesizer Developer forum? I'm willing to report it. (Although my time is stretched trying to get this course reviewed and fixed by the deadline.) I'm not against waving problems under the noses of the right people to fix them--hooray for that!, but I will admit to feeling a bit exhausted as the sheer number of problems-that-should-be-fixed (specifically when trying to make a 600 page math course accessible) is daunting.





On Fri, Aug 19, 2022 at 3:11 PM Brian Vogel <britechguy@...> wrote:
On Fri, Aug 19, 2022 at 02:49 PM, Joy Holly wrote:
Thank you for the breadcrumbs leading me to the information I needed most, which was essentially confirmation that the math is accessible, and the reading issue here lies with NVDA specifications that need tweaking.
 
-
Joy,

I am not trying to beat a dead horse here, as I've been where you are myself.  You can say, accurately, that an NVDA setting for voice may need tweaking, but it is critical to note that the issue does not lie with NVDA (nor JAWS, nor Narrator) but with the synthesizer in use.

There is nothing for the developers of NVDA (or JAWS, or Narrator) to fix in an instance like this.  A report could be made to the developers of the wonky synthesizer so they are aware of this odd bobble when reading mathematical expressions, but it would be up to the synth developers to fix it, not the developers of the screen readers that can send text to that synth.

This is not as hair-splitting as it seems, as getting issues under the noses of the right people to fix them is the quickest way to get them fixed (and that's only if the brains above those noses care). 
--

Brian - Windows 10, 64-Bit, Version 21H2, Build 19044  

The difference between a top-flight creative man and the hack is his ability to express powerful meanings indirectly.

         ~ Vance Packard


 

Joy,

In Windows 10 and 11, the Feedback Hub is the very best way to report issues, and fairly easy, too.

--

Brian - Windows 10, 64-Bit, Version 21H2, Build 19044  

The difference between a top-flight creative man and the hack is his ability to express powerful meanings indirectly.

         ~ Vance Packard


Joy Holly
 

Thanks! I will put it on my to-do list. It's definitely an annoying reading bug.


On Fri, Aug 19, 2022, 4:32 PM Brian Vogel <britechguy@...> wrote:
Joy,

In Windows 10 and 11, the Feedback Hub is the very best way to report issues, and fairly easy, too.

--

Brian - Windows 10, 64-Bit, Version 21H2, Build 19044  

The difference between a top-flight creative man and the hack is his ability to express powerful meanings indirectly.

         ~ Vance Packard


 

Hi,

Hopefully one last thing: I think it might be helpful to tell accessibility engineers (at Canvas) to inform technical authors and curriculum writers about things to be aware of when using screen readers to access the online content (hopefully the docs would be written from the perspective of the student/teacher/parent/support specialist/anyone willing to review the course materials and the tech implementation used; if folks need guidance on wording and such, feel free to contact Brian, I, and others who are either instructors or have teaching experiences, especially folks with assistive tech experience).

P.S. I might as well save this conversation somewhere as my university uses Canvas for LMS (learning management system).

Cheers,

Joseph


 

On Fri, Aug 19, 2022 at 04:39 PM, Joy Holly wrote:
I will put it on my to-do list. It's definitely an annoying reading bug.
-
Joy,

Another one of those seemingly hair-splitting things that probably isn't important here, but could be:  The Feedback Hub in Windows 10/11 can only be used to communicate with Microsoft regarding a synthesizer that they maintain (or anything else you can possibly think of that Microsoft maintains).

I'm saying this more for the historical record than anything else.  If the synth that's misbehaving is issued by some other software house, then their technical support is where the cage-rattling needs to occur.
--

Brian - Windows 10, 64-Bit, Version 21H2, Build 19044  

The difference between a top-flight creative man and the hack is his ability to express powerful meanings indirectly.

         ~ Vance Packard


Brian's Mail list account
 

OK, but assumedly you are using at least Windows 10, so NVDA will default to one core I think I'm right in saying. though it may not suit everyone, try changing the synth to Espeak and see if there is a change. Sometimes Microsoft voices tend to assume too much and this can screw things up. I do not know what voice is default for Jaws these days, but its still probably Eloquence, which like Espeak is a non human voice and therefore can say most things correctly.
Brian

--
bglists@...
Sent via blueyonder.(Virgin media)
Please address personal E-mail to:-
briang1@..., putting 'Brian Gaff'
in the display name field.

----- Original Message -----
From: "Joy Holly" <sjoyholly@...>
To: <nvda@nvda.groups.io>
Sent: Friday, August 19, 2022 7:49 PM
Subject: Re: [nvda] Math Reading Issues (pronunciation of x^3, y^3)


Want to preface by saying in no way was I hoping anyone here would do
troubleshooting I can do, was just looking for feedback to the question "is
this sort of math reading error common/known" so I could rule out it being
something with the LMS, with the version of Mathjax used, etc. to widen my
knowledgebase before knowing what to try out next.

I will now answer my own question, being given the hint by you all to
experiment with the set synths and voices. I had not done this because the
goal on our side is to provide accessible and readable math, with the
responsibility for the customization needed for this screenreader or that
being with the user.

This reading error occurs also in FF, and I should have said this (but
Chrome is our target browser since it is the one used by the schools.)

The default synth setting is what we are using since we adjusted nothing
after installation. It is Windows OneVoice with the voice Microsoft David.
I tries the other two OneVoice voice options; the pronunciation issue
remained. So then I changed the synth setting to Microsoft Speech API v. 5,
using the Microsoft Dave voice, and now it reads "x cubed" as expected!

Thank you for the breadcrumbs leading me to the information I needed most,
which was essentially confirmation that the math is accessible, and the
reading issue here lies with NVDA specifications that need tweaking.

Very grateful!

Joy

On Fri, Aug 19, 2022, 1:50 PM Brian Vogel <britechguy@...> wrote:

On Fri, Aug 19, 2022 at 01:42 PM, Rui Fontes wrote:

To you "eekup" seems an english message?

For me it is not an english message...

-
Which is precisely the point and the problem. It's isolated to cubed
variables, and one of them, the X one, is at least far closer to what it
should be than the Y one is.

All other instances referenced make it crystal clear that the announcement
is both correct, and in English.

I don't know how that could possibly have been clearer from the initial
problem statement, message #98716
<https://nvda.groups.io/g/nvda/message/98716>. This is one of the
best-written complex problem statements I've ever had the privilege to read
on this group.
--

Brian - Windows 10, 64-Bit, Version 21H2, Build 19044

*The difference between a top-flight creative man and the hack is his
ability to express powerful meanings indirectly.*

~ Vance Packard