Comment 1 by jteh (in reply to comment description) on 2013-08-07 04:08
Replying to soiffer:

One obvious thing is that NVDA should only ask for the math once, not three times. The three requests come because NVDA asks for the acc_value, acc_name, and acc_description.

We ask all MSAA objects in IE for this because it is usually different. We'd need to special case for MathPlayer here.

What would be even better though would be to avoid querying the name/value/description for every (math) element on the entire page during loading.

It'd be nice if we didn't have to pull the content for the entire document, but it's the only way we can provide a flat representation of the document with fast, consistent navigation.

Comment 3 by soiffer on 2013-08-07 05:25
A special case for a node named "math" sounds like a good idea. That would cut the time down by 1/3. If you do a special case, I do highly recommend trying to get the MathPlayer interface and passing in the TTS associated with the voice along with the current rate, pitch, and volume (the rate is the most important). You will get much better speech out because then the speech embeds higher quality prosody info into the text it returns and it will sound much more intelligible/better.

You don't need to assume MathPlayer is there, simply check if it is there and if so, make the four calls. I have sample code in C++ if that would help.

Comment 4 by jteh on 2013-08-07 05:57
Our browse mode doesn't deal with raw speech or braille, just content, so retrieving raw speech doesn't work for us. Even if we added some hack to push raw speech, we'd run into problems when the user tries to move through it word by word, character by character, etc. Also, some synths don't understand SAPI 5, SSML, etc.

Comment 5 by soiffer on 2013-08-07 06:16
Perhaps you misunderstood. MathPlayer just returns the text to speak. But if you tell it the standard supported by the TTS engine (SAPI 4/5, ECI, SSML), then it will generate text with embedded speech cues that are understood by the speech engine. If you store the raw speech text that includes those cues, then indeed navigation could potentially be affected by these cues (e.g., )

Other AT has dealt with this. I think they do this by simply skipping over these commands when navigating. Of course, you would need a flag on a text run that said that the embedded cues need to be skipped, but the syntax is well defined for the cues and I think the code to skip over them is relatively straightforward. I'm sure that this would take a day or so to code, but the result of having good prosody is much more intelligible speech that is much easier to listen to.

Comment 6 by jteh on 2015-07-09 23:42
No longer valid, as we don't support the old IE MathPlayer <= 3 any more in favour of the new MathPlayer >= 4.
Changes:
Added labels: invalid
State: closed