Tuesday, August 24, 2010

Transferring a Dragon NaturallySpeaking User Profile Between Computers

Note: This article applies to version 10.1 of Dragon NaturallySpeaking.
The overall process for transferring your user profile between computers involves exporting the user profile from computer A and importing it into computer B. The process for doing this is described in the online documentation (DragonBar Help menu -> Help Topics menu item). Below are the steps with some added commentary.

To Export A User (Computer A)

  1. On the NaturallySpeaking menu of the DragonBar, click Manage Users.
  2. In the Manage Users dialog box, select your user and click Advanced button, then click Export.
  3. Use the Browse For Folder dialog box to select a location for the exported user files (e.g. a flash drive) and click OK.
  4. When the export operation completes (it may take several minutes), close the Manage Users dialog box.

To Import A User (Computer B)

  1. Start NaturallySpeaking. When the Open User dialog box displays, click Cancel button (you do not need to have a user open to perform the import).
  2. On the NaturallySpeaking menu of the DragonBar, click Manage Users.
  3. On the Manage Users dialog box, click Advanced button (you do not need to have a particular user selected), then Import.
  4. Use the Browse For Folder dialog box to locate the user files you want to import (you need to select a folder; the export process creates a folder whose name matches the user profile you exported, so look for that on your flash drive) and click OK.
  5. If the user profile already exists, you'll be asked if you want to overwrite it or import it with an alternate name. If you choose the latter option, you'll be able to modify the user profile name in that dialog box.
  6. When the import operation completes, close the Manage Users dialog box.
  7. To open the user, on the NaturallySpeaking menu of the DragonBar, click Open User.
  8. You'll be prompted to run the Audio Setup Wizard so have your microphone plugged in. This only takes a minute or so. After this you'll be able to dictate.

DNS Versions

When I first tested this process, I exported from and imported into the same DNS version (10.1). After all, the idea is that you are transferring your user profile between multiple computers upon which you have DNS installed (see my article on Installing Dragon NaturallySpeaking 10.1 on Multiple Computers). If for some reason the version of DNS on both computers are not the same, I hypothesize that you won't be able to import your user profile into an earlier version of DNS but you will be able to import it into a later version. This is based upon a DNS Knowledgebase article that says, "While the user profiles can be used without issue on a higher product version, they are not backward compatible." To test this, I imported my 10.1 user profile onto a computer that's running the original version of my software, 10. While the import was successful, when I tried to open a user, I received the following error message: "Com returned an unexpected error code: Details are E-FAIL."

User Profile With Multiple Dictation Sources

If your user profile has multiple dictation sources assigned to it, all of them will be exported and imported.

Achieving Similar Recognition Accuracy Across Multiple Computers When Using A Second Computer Infrequently

If you use another computer (call it computer B) infrequently and want to achieve the same recognition accuracy as you have on your primary computer (call it computer A) after you've just imported your user profile, you will come closest if you use the same microphone in both places.
Recall that DNS adapts to both your voice (acoustic model) and your  vocabulary (language model). Also, recall that a user profile can have multiple dictation sources assigned to it (e.g. standard wired  microphone plugged into sound card, USB microphone, Bluetooth, digital  voice recorder); this is desirable because then they all share the same language model (vocabulary). Finally, recall that while your  acoustic model depends on the microphone that you use, other factors also contribute to it such as the type of sound card (if you aren't using a USB adapter), how noisy the computer is, background noise level and type.
What this means is that the acoustic model is not going to be a perfect match when you transfer your user profile to another computer. But the more variables you eliminate (e.g. mic, sound card), the more similar the acoustic models will be.
If for some reason you cannot use the same mic on both computers, you might want to add mic B as a new dictation source to computer A. You'd still use mic A with computer A but the advantage of this is that you wouldn't have to go through the extra step of adding mic B as a dictation source after the import on computer B; this process takes approximately 10 minutes because you have to run the Audio Quality Wizard (volume check, quality check), Training Wizard (short training takes 5 minutes) and User File Adaptation and Save (about 2 minutes). When you export a user profile, all dictation sources come along with it. So if you already have the mic B source created, you just have to run the Audio Quality Wizard (about 2 minutes) before you start dictating.
A disadvantage of using mic B only with computer B is that mic A is the one you're constantly using so DNS is improving its acoustic model but these adaptations aren't taking place for mic B. I'm not sure how much of an issue this might be for your particular circumstances. When using mic B, you still have access to your latest-and-greatest language model which includes custom vocabulary that you've added as well as statistics such as word frequency and proximity to other words.
So I recommend that you run an experiment to gather some empirical data as follows:
  1. Find text that is representative of the type of dictation you will be doing on computer B; save it in electronic form in Microsoft Word as original-text.doc
  2. On computer A
    • add a dictation source to your user profile using mic B
    • save your user files
    • export your user profile
    • dictate for several pages (2-5) into Microsoft Word using both micA and micB; save these to two different files (computerA-micA.doc, computerA-micB.doc)
  3. On Computer B:
    • import your user profile
    • use both micA and micB to dictate the same text; save these to two more files (computerB-micA.doc, computerB-micB.doc)
  4. Compare the results with Microsoft Word's Compare feature i.e. analyze the differences.
    • assess recognition accuracy for your main user profile (primary computer)
      • compare original-text.doc with computerA-micA.doc
    • assess similarity of recognition accuracy for your preferred secondary-computer setup
      • compare computerA-micA.doc with computerB-micB.doc
This will give you a rough idea of how accurately DNS recognized your dictation when you performed it on your primary computer.  The second comparison will show you the differences in your dictation when using the 2 different computers with the two headsets. If the differences are minor, you can feel confident that using a different microphone with computer B won't have much of an impact on recognition accuracy.
If the differences are major, compare the dictation when you used mic A i.e. computerA-micA.doc with computerB-micA.doc. If these differences are also major, then some aspect of computer B or the environment it's in are impacting the acoustic model.

No comments:

Post a Comment