Bug: 2488167
Don't show ko, ja and zh languages in the list.
Remove a debug println.
Add alternates_for_g to the keyboard for it to work on turkish. This
must have not gotten merged from donut.
dictionary.
Example, "ho9" would match "Shoemaker", if "Shoemaker" existed in your
contacts. This was due to premature switch to completions mode when
trying to insert a missing letter.
This syncs up the Java version to what the native one was doing.
This change seems to fix it. Will continue to monitor if it happens.
From tracing the code, it seemed to think that the current keyboard
was not a symbols keyboard, so the "equals" call was failing. It was
probably due to voice mode changing between keyboard creation and
display.
This also addresses the defaults for voice button on a wiped device.
I think it also addresses mic button showing up when not expected by a specific text field
that has the privateImeOptions of "nm".
Bugs 2417842, 2242893
Due to the addition of input language switching, the auto dictionary
will be re-created very frequently. We need to save it and restore it
during transitions, otherwise the data will be purged too often to be
of much use in the multi-lingual case.
This also fixes the case where a user frequently turns the phone off/on.
Auto dictionary data was not being saved in those situations either.
The dictionary will probably never grow too large since most of the
frequently used words will be moved to the user dictionary any way.
Also, now the input locale is different from the display locale, so
save the input locale in the UserDictionary class and insert words
into the user dictionary with the correct locale.
And NPE fix for bug: 2464573
The last change did not solve the bug completely, the spinner was still shown at times when no initialization dialog was shown
Change-Id: I43a5f8df4e333ffb6deb727c3bf45ed7b7a54069
This increases the chance of hitting the correct letter when typing a word
that exists in the dictionary, rather than only correct it after the fact.
It is most effective after 2 or 3 letters of a word have been typed and gets
more accurate with more typed letters in the word.
If 2 adjacent letters have similar probabilities of occuring, then there is no
hit correction applied.
Now there is a delay each time the voice recognition screen is opened, however, this is the correct behavior as microphone setup takes about 600ms.
The previous implementation, although being fast, was not actually recording the first 1/2 second of user's speech.
Change-Id: I8b8373a59658502f7451150ff4a4ffd1c8e83205
