Dragon Naturally Speaking

What works

Dragon NaturallySpeaking 10 Standard is now very useable in continuous text dictation mode.
It installs without a problem, it trains well, it does not crash. Dictation into DNS's DragonPad and DictationBox, Wine's Notepad, and Notepad2 work very well.
Not all of its features work, but not all of its features are helpful to the average user, either.

The DragonBar shows up on top, as well as the tray icon. The DragonBar can be set to cling to the top of whatever program it is dictating into. Sadly, the buttons do not show up, but all the content can be viewed by mousing over the DragonBar.

While dictating, a few of the verbal commands, such as "Correct that," "Scratch that," etc. work, as well as most of the grammar commands ("Cap that," "hyphenate that" etc.) work when dictating into DictationBox and DragonPad. In notepad, the command is recognized as a command but does not execute, and quite often the program freezes.

DictationBox has a feature to make its text buffer non-volatile, which means that text can easily be pasted into Linux application using CTL-V. This is a very useful feature, since DNS's continuous text entry is fastest and most accurate when using this environment, and most users seem to prefer it. To access this feature, open DictationBox, Settings, and then mark "Keep Transferred Text in Buffer." Make sure that the CTL-C and CTL-V options are checked.

I use the program as follows: open DNS, open wine's Notepad, open DNS's DictationBox. Dictate into DictationBox. Say "Transfer." At that point, text transfers both to Notepad and to the internal buffer. I then mouse over to the Linux program I am working on and do CTL-V. The text drops, all but the last 2 characters.

The Accuracy Center works once you install Gecko, but not all the commands work within it. Of the useful menus: audio settings, acoustic optimizer, and the vocabulary optimizer run well. 

What does not

The most annoying problem is that the drop-down menu headings on DragonBar are invisible. The buttons are there, but you can't see them. When you mouse over them to see the submenus, an error box pops up, telling you that an internal error has occured. Just close it. Some of the drop-down menus do not work. The language model optimizer does not, for instance.

In addition, there is some problem between the linux sound system and wine. To run the program (and many wine programs) you cannot be using pulseaudio. I have purged pulseaudio from my machine. You may wish to use pasuspender or pasdp. You will also need to set winecfg to use OSS only. Directions on how to do this are available elsewhere.

Because DNS only works within the wine environment, it cannot enter text into non-wine applications. This is not as big a handicap as you might imagine. See the above instructions to use DictationBox and CTL-V. 

Workarounds

What was not tested

Some of the minor features, user help features. Languages other than US English.

Hardware tested

Graphics:

• GPU:
• Driver:

Additional Comments

Dragon NaturallySpeaking is a wonderful program when it runs. There is nothing like it. Its great virtue is that it can enter reams and reams of text quickly and accurately, with no strain on the fingers. It is invaluable to writers, researchers, transcriptionists, and plain old bureaucrats. In many parts of the US, and overseas, DNS is being used for some court reporting. And when DNS is running well on Linux, its speech recognition engine runs GREAT. Very fast. Very accurate. One problem users have is with sound quality. To achieve really good results, a better microphone is needed than the one supplied with the program. And user training is essential. Users must speak in a non-noisy environment, in a steady tone of voice, and very clearly. When I first train the program, I look for an accuracy score of 25 or above, even though the program will run with less. With practice, you can speak VERY quickly. In addition, DNS in any environment is a resource hog and works best with a fast machine, lots of RAM, and not a lot of other programs running in the background. It also may require exclusive access to the sound card, depending on your configuration.

There are more complete installation and setup notes on the version 7 entry, many of which will also apply to v10 though expect some differences due to changes in WINE as well as DNS. 

 This should however be a useful reference.

 The following notes were contributed by Gurdil to set up DNS 10 and Wine to work with Pulse Audio on *buntu . Not tested elsewhere but may well have broader application.

padsp command is the key and is pulse audio specific.

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

sh winetricks fontfix fontsmooth-gray gecko sound=oss gdiplus

2. Insert the installation CD then copy contents to the hard disk, ~/DNS for example

3. cd ~/DNS/issetupprerequisites, run "wine vcruntime.exe"

4. cd ~/DNS and run "wine setup". Proceed with the installation. At the end choose to register with printing something

5. run "padsp winecfg". Go to audio and choose alsa.

6. run DNS with: "padsp wine ~/.wine/drive_c/Program Files/Nuance/NaturallySpeaking10/Program/natspeak.exe" If it crashes, run it again.

7. Now it should work. If it does not detect your microphome (or explains that sound is too low), try to configure your input device (menu>sound and video>pulse audio control)

Some editing commands work in notepad, and the recognition is still almost as fast as in DictationBox. For right now, I recommend dictating directly into Notepad.

The program no longer freezes when using Notepad. It sometimes still does after using DictationBox extensively. For instance, I dictate 4-6 paragraphs at a time, and after about the third session of that, the program sometimes behaves strangely. Usually what happens first is that the command "Transfer" does not drop text automatically into the other program.

Solution, just go to the other application -- be it wine or Linux -- and hit CTL-V and the text will drop.

Interesting note, unrelated to Wine... Dictation accuracy changes when switching from Notepad to DictationBox. Usually drops a bit at first.

These instructions are primarily for Ubuntu / Debian but should be helpful to other distros. 

In terminal:

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

sh winetricks fontfix fontsmooth-gray gecko sound=oss

(also add gdiplus if you want a prettier DragonBar.)

winecfg

(in winecfg, go to audio tab and check oss driver only, then bump the default sample rate to 48000)

navigate to CD rom and type

wine setup

When evaluating install-menu options, remember that the program will need to run simply. Disable options that require searching your hard drive, automatic updating, sending data to Nuance, etc. Don't install or enable training.

The way to get fast, accurate dictation is to use Notepad or DictationBox. DragonPad works and has some extra commands but it is not as fast. It also has some quirks. Default font cannot be changed, for instance.

Once you have the program up and running, you should be able to use Dictation Box, say "Transfer that" and then use CTL-V to paste into any running Linux application. To enable this feature, open Dictation Box / Settings, and check the box to keep transferred text on the clipboard.

Note: I compiled wine from git using ./configure --enable-win64 and NatSpeak ran just fine. (However, check your Natspeak version. Only 10.1 is 64-bit enabled.)

Additional Instructions:

As of 10.10 Ubuntu, pulseaudio is enabled by default. I have not had luck getting wine programs running with pulse. To disable it, edit /etc/pulse/client.conf to change # autospawn = yes to autospawn = no.

Then System/preferences/startup applications, uncheck everything that says pulse. Then reboot.

If you want to use a USB device, this must be set as the default device. So add to your home directory a file called .asoundrc that contains the following text. (This assumes that you have only two devices, that card 0 is your on-board and card 1 is your usb.)

pcm.!default {
    type hw
    card 1
}
ctl.!default {
    type hw
    card 1
}

It is my experience that the Nuance licensing daemon now counts each new version of wine as a "new operating system." Upgrades between, say Ubuntu Karmic and Ubuntu Lucid, also use up a license. And you also lose a license when you have NatSpeak installed twice on the same computer.

I had been installing test versions daily, always on the same machine, always using the latest wine git, and now can no longer do so. So new users may not want to install or test casually. 

Note: Changing the Windows Version in winecfg counts as an operating system upgrade. For this reason, it is recommended to install DNS10 under an alternate $WINEPREFIX and leave it alone. Nuance recommends un-installing the software with an active internet connection before upgrading. This is supposed to prevent the activation count from incrementing. The "active internet connection" seems to require the optional winbind (samba-winbind) package available through your distro. If the license loss continues after un-installation, bug report it. http://nuance.custhelp.com