Dragon Naturally Speaking

What works

Installation and training. Basic dictation into DragonPad or Dictation box.

Some voice commands such as "scratch that" "go to sleep" (mike off command).

What does not

Accuracy centre. Voice play back crashes.

Workarounds

What was not tested

Tutorial and text-to-speech were not installed nor tested. Vocal cursor movement commands not tested.

Hardware tested

Graphics:

• GPU:
• Driver:

Additional Comments

Installation requires some fiddling but does work fully. See notes below.

The following text was dictated using Dragon 9 Pro. There are only 2 or 3 very minor faults.

This is my shiny new installation of Dragon naturally speaking on Linux.  It required running the install process three times.  The first installation was done with default settings. It finally fails just before the end leaving all the programme and tutorial files in temporary directories.

The second installation allowed customisation of the install destination and language options. This installation gets almost to the end and then reports unexpected termination.

Finally a third installation process with the same customise options is the last step allows it to get right to the end.  Dialog boxes shown confirming completion of installation process.  At this stage the programme can be run under wine and enters into the setup for the voice training.  Once a voice training is completed there is a small delay while its doors [it stores] profile information, then the programme is ready for dictation.

­
rm -rf ~/.wine
cd ~/dl

wget http://www.kegel.com/wine/winetricks

sh winetricks win2k gdiplus sound=alsa vd=800x600
cd ~/.wine/dosdevices
ln -s /mnt/tmp h:

# disable internet access or else it faults on some dlls: winsta.dll regapi.dll

wine h:setup
# keep std options, install
# copying slow, be patient.
# dns "error: auth data (44952) not properly read or written."
# dns "couldn't find reg file ... "
# deselect option : "check internet for updates"
# terminates correctly to command prompt but desktop window remains.
# all dirs in  ~/.wine/dri*/Prog*/Nu*/* are named like *temporary9

wine h:setup
# install shield: modify / repair ?
# choose modify:  take out text-to-speech, tutorial and "foreign dialects"
#                 keep UK (or other required)
# reports exits unexpectedly but controlled, not crash.
# terminates correctly to command prompt

wine h:setup
# install shield: modify / repair ?
# choose modify:  take out text-to-speech, etc, as before.
# exits cleanly with final confirmation dialog.

cd ~/.wine/dri*/Prog*/Nu*/*/Prog*
wine natspeak

­

 It will now start to configure initial user profile and enter voice training.

Once done it will save the profile. Skip document analysis , start with dictation. Let it save the profile then close the user. 

Now to turn off yet another phone home feature that may bring up dll problems. Second menu on the main menu  has "administrative settings", go to Miscellaneous , deselect "check for product updates". Close the program to ensure all is saved and stable.

On restarting, It will take some time to load the profile when restarting Dragon so don't presume it has crashed. Hover mouse over the mike icon in the task bar will show a hint message that it is "initializing dictation..." .

This method was used with WINE 1.3.8 on Gentoo Linux which provides Wine with gecko , many distros do not. You may need to use the distro to add gecko or add it to the winetricks options. This was not tested here.

Despite the triple install loop it does work well, I was able to install the options I required in non std install directory. Just don't try it first time in. :?

To the Mandriva user who could not get Linux running in Spanish. 

I have a pre-beta of Ubuntu's Intrepid Ibex and the most recent Wine git. I have installed and and am running DNS 9.0 preferred. It runs well.

However, having said that, I did get the problem you described, once or twice, both with the latest Wine. I tried to reproduce it so I could file a bug report, but could not. Please file a bug report in Wine with all the information you can.

Thanks.
Susan Cragin

Also see note below on how to install v 9 Pro on more recent Wine versions. This howto is somewhat old and may not apply to recent Wine releases though tips could be useful.

These are the current installation instructions for WINE rc1, plus some tips on training the software and running the program. 

Installation instructions:

wget http://www.kegel.com/wine/winetricks

sh winetricks fakeie6 gdiplus

(hit the enter key)

(copy the contents of both DNS 9.0 preferred disks to a "DNS" file, and navigate to it.)

wine setup

(follow installation instructions.)

When you have finished installing the program, click on the Main Menu icon to start the program. 

Hit "Activate Now."

Obtain a product activation key by clicking the link. Save the number in a secure location and then enter it manually.

TRAINING (Most of this is good advice for Windows users too.)

Follow the training instructions as given. During the volume check and accuracy check, pause for about a second before speaking, so the machine knows what "silence" sounds like. If your accuracy number is below about 19, you may have a poor-quality sound card or your Linux distribution may not be recognizing your card correctly. See the help desk for your distribution if this occurs.

For training, select either Kennedy's speech or Alice in Wonderland. The other choices do not produce as good voice models.

Sometimes the windows do not seem to work properly. Just hit "back" or"pause" or "next," and everything will be fine. There is a date in the Kennedy speech that often halts recognition. Just hit "next" if the halt occurs.

RUNNING THE PROGRAM

Dictate into the included DragonPad if you want to activate the maximum amount of commands. Dictate into Notepad if you want speed and accuracy. Both operate only within their own windows. 

When you are finished dictating, the program will ask if you want to save voice training data. Do not do this. WINE saves the training data accurately, but this is not a good idea because the saved data, over time, degrades the voice model. 

Other options to enter text are Notepad2 and NoteTab. Notepad2 has a nifty keystroke combo to move paragraphs of text around. (These programs offer free versions for download.)

But remember, the more complex the program you use, the slower and less accurate will be the text entry.