Screen Reader NVDA Japanese

Fork

• howto
• nvdajp
• nvdajpmiscdep

Added a 'say cap before capitals' setting to the voice settings dialog, this can be used for synths that don't change their pitch, like some of the Reap Speak synths. Got rid of relativeUpercasePitch setting, its just 25 all the time.

Replaced the voice dialog (that could only change the voice) to a proper voice settings dialog that can allow you to change the voice, rate, pitch, volume and punctuation status. All controls work in realtime, if you cancel, all changes are reversed, if you press OK all changes are kept, though you still ned to save the config before you exit NVDA for the changes to perminantly exist. Preferences Menu is renamed to speech menu.

Bullet and numbering fixed in Mozilla Gecko virtualBuffers. The mozilla list item object now grabs the bullet and uses it as its name, also it removes the bullet from its list of children.

Documentation updates, ignore graphics in Mozilla Gecko virtualBuffers with no title or alt attribute, page down and page up in rich edit documents now speaks the line moved to.

*Made loading of documents in Internet Explorer virtualBuffers slightly faster by murging getDomNodeInfo and getDomNodeText all in to fillBuffer, which cuts down duplicate calls to properties on an MSHTML domNode object.
*Finally completed support for dynamically updating documents in Mozilla Gecko virtualBuffers. Previously most of the code was there but it had some serious bugs which would completely mangle some pages. Now using the idea of z ordering (and isn't it great how you can compair python lists... [0,2,4] is less than [0,3]), pages are all kept in the right layout, and it seems now that all examples at
http://accessibleajax.clcworld.net/
work fine.
*Changed the default value of config option updateContentDynamically in virtualBuffers section to True. Previously it was false.

text is now checked for characters that appear more than 5 times in a row, and they are shortened to 5. Example: a line of dashes will be only spoken as 5 dashes '-----'.

Fixed bug where negative states were being reported as positive states.

Made sure that the first lot of new text is spoken after pressing control+c in dos consoles. Also added an api.getDesktopObject and api.setDesktopObject. Eventually this will be used when the foreground and focus object get hidden.

More fixes to dos console new text reading. speakNewText function is now called calculateNewText and it returns the new text. MonitorThread uses this to get the new text and speaks it only if its clear it isn't a character that has been typed. However, it is much more correct now with this and doesn't block new text that cleary can't have been a typed character.

*Fixed bug in sapi5 where it wouldn't shut up for quite a while, even if the control eky was pressed, because too much text was sent to the synthesizer. (e.g. big lots of text in a dos console). Cancel now sends the purgeBeforeSpeak even if the synth doesn't think it is speaking. This doesn't seem to break any of the sapi5 synths that I can currently test.
*Improved the reading of new text in dos console windows yet again, by now using both the MSAA console events and a monitor thread. The Monitor thread waits 100ms after the first event and then tries to speak new text, rather than trying to speak on every single event.
*Trapped an exception that sometimes can happen in the windows API waitMessage function.
*Removed the playing of the question sound when bringing up the NVDA exit dialog. I'm hoping that this is what was causing NVDA to freeze when pressing insert+q sometimes.

*NVDA no longer crashes when pressing control+c in dos console windows. Fixed by adding a control+c script which disconnects from the console, sends the key, then reconnects.
*Loading documents in mozilla gecko virtualBuffers (e.g. Firefox) now takes less than half the time it used to. Example: www.anz.com used to take around 8 seconds, now it takes only 3. Fixed by massivly cutting down the time it takes to initialise an IAccessible NVDA object (it doesn't call windowFromAccessibleObject unless it really has to). Also all hash calculations for NVDA Objects has been moved in to __hash__ rather than calculating on initialisation and then making __hash__ used the cached value.

Made the beep speech mode pitch configurable. beepSpeechModePitch, in the speech section of nvda.ini.

*the tones module now caches the waves it generates in the waves directory as "_tone%sHZ%sMS.wav" where the first %s is the pitch in HZ and the second %s is the length in milliseconds.
However to avoid filling up disc space, the files are deleted each time the tones module is loaded (i.e when NVDA starts).
*insert+s used to turn speech on and off, but now it toggles between 3 speech modes: talk, beeps and off.
Talk is just normal speech, off is off, but beeps, is an idea that came from the Dos screen reader, ASAP. This mode plays a 5ms 10000HZ tone (very high and very short) each time an item of speech is supposed to happen. It is very responsive, and it allows you to deal with the situation when you are doing something that is constantly changing the screen, as in millions of files are scrolling up a dos console, or perhaps you are running a setup program that is constantly telling you what file it is copying etc. If you switch to the beeps speech mode, you only have to wait until the flood of beeps stop and you know the task is finished. This is much faster than having to keep pressing control, or waiting for all the speech to finish.
*Shortened the length of progress bar value beeps from 100MS to 40MS.
*Changed the speakNewText function for dos consoles to collect together all new peaces of text and send them to the speech synth all in one go (separated by spaces). Previously each time a peace of new text was identified it was sent.

Catch an exception if a rich edit control doesn't support page numbers. My Outlook Express seems to be doing this at the moment.

Some more documentation updates, also now empty lists say 0 items when given focus. Previously they sometimes were saying 1 items or 2 items, because of their list headers being counted in the child count.

Changed a few default config options: reportLineNumbers is now False. Fixed bug in the speakCommandKeys code that was either now allowing it to speak the actual key, or it would say the modifier keys and then some <function blablabla> junk.

Documentation updates, script_navigator_reviewMoveToCaret re-added as numpadSubtract, and tones amplitude is now 10000 rather than 15000

Changed the startup message to say 'NVDA started' and renamed the title of the NVDA window to 'NVDA Interface'

Updated setup.py to copy the waves dir to the distrobution when compiling

Converted sound files to 22050 mono

*Sound files (in the waves directory) are now used in NVDA to denote events such as startup, exit, error and crash. Previously the PC speaker was used.
*Beeps that indicate the value of progress bars are now generated and playd as a sound file. Previously the PC speaker was used.

Fix for exceptions on focus; was experimenting with the mouse following the navigator, but it won't work due to menu itmes doing strange things when the mouse is over them.

Fixed a few exceptions in outlook express and internet explorer.

*Mouse shape changes are now reported. (to turn off, set reportMouseShapeChanges to False in the mouse section of nvda.ini).
*When mozilla gecko virtualBuffers are loaded, say all is no longer automatically performed. If the pages was for some reason loaded straight after a load, or there was a reorder that caused part of the page to be updated, say all would majorly break and then crash Mozilla gecko. Now it just says the line that the cursor is at instead. This has made it a bit safer to use updateContentDynamically in mozilla gecko virtualBuffers.

Added support to the mozilla gecko virtualbuffer code to dynaicmailly update content on event_reorder. This means that changes to a page due to javascript etc should be now shown. Also the updating of form fields is also now supported. However, on large pages this sometimes can crash the web application plus NVDA so to use this code, updateContentDynamically needs to be set to True in nvda.ini. This is False by default.

Optimised code a bit, gecko virtualBuffer is 3% faster at loading documents due to murging getNVDAObjectInfo and getNVDAObjectText in to fillBuffer, this cut down on calls of NVDAObject properties. Re-wrote a few for loops as maps, according to some people's experiences, map seems to be faster than a full for loop. Also changed range to xrange in a lot of places.

Many little fixes to problems reported by pyChecker. Also changed a few map calls to 'for' loops because some of the calls were only changing some of the items, therefor its better not to use map.

Reporting the object under the mouse as it moves can be now turned on or off by insert+m

Added javaAccessBridgeHandler.py. So far this only contains an initialize function, plus ctypes structures of all the major java access bridge structs needed. This is in no way yet connected in to NVDA properly.

Fixes to setup.py and comtypesClient.py so that compiling to an executable works again.