Notes from a Linguistic Mystic

So, it’s now been about a month since I started using Dragon Dictate 3, and I thought I’d give a bit more discussion based on long-term use following my original review. I won’t bother commenting on the things that haven’t changed at all with additional usage, but some things have really jumped out at me as I used the software more.

Use a headset mic

The biggest single thing that I did to improve my experience with Dragon Dictate was to order a real microphone. I had been using a very old headset microphone which came with a copy of IBM Via Voice around 2000, or, when I was away from home, just using the built-in microphones on my computer. This worked well, especially the built-in mics, but I found myself growing a bit annoyed by having to disconnect my computer from the external monitor and open it up in order to do any kind of dictation, so I went online and ordered an Andrea NC-185VM Headset Microphone, which is on Nuance’s recommended hardware list.

In a word, wow. I was getting relatively good accuracy with the built-in microphone on my computer, but moving to an actual headset mic, and to a modern one, really improved the accuracy I’m getting. More importantly, it also gives me the ability to stand up and walk around a little bit as I dictate. This is a small luxury, but an appreciated one. So, if you don’t have a headset microphone, it may actually be worth ordering the software as a physical copy with an actual headset, and if you did the digital download, this may be a good time to upgrade your headset. Even as a speech geek, I’m impressed with how much this was helpful.

Training and vocabulary

One thing that I’m very impressed with is the effectiveness of the training process. I use Dragon to dictate relatively intense linguistic documents, using obscure words like phonology, vowel, monophthong, nasality, and assimilation. I dictated that whole sentence, and didn’t have to correct Dragon at all. I’m very impressed with its ability to learn new vocabulary, especially technical vocabulary, and implement it very quickly.

Also, I really do feel like this software is learning effectively. Unless there’s a major increase in background noise or something along those lines, I feel like every dictation session is subtly better than the last, and there are relatively few errors, especially with the headset mic, that occur over and over again.

Phonetically Predictable Troubles

As a phonetician, it’s always interesting to examine the kinds of errors that you see over and over again in this software. for me, the most consistent error is the substitution of the vowel /æ/ (as in “pat”) from /ɛ/ (as in “pet”). Acoustically, for me, those vowels are very similar, so it’s not shocking that Dragon struggles with differentiating them. Similarly, it has trouble distinguishing “Will” from “Well”, especially frustrating because both are high-frequency words, and one of them’s my name. However, as somebody who studies vowel perception, I understand that there is no good way to automatically make that distinction, and that most of the time when humans do so, we’re simply using context to guess the best candidate.

Stupid text entry errors

My biggest complaint about Dragon Dictate has nothing at all to do with speech recognition. For whatever reason, when dictating into the Mac, every so often, Dragon will omit letters. In the penultimate sentence, Dragon ended it with “with speech recognitio.”, completely dropping the N. This happens very frequently at the ends of sentences, not always with N, but often with the letters T and I. similarly, it can get itself into a mode where adding new text overwrites the last word of the prior sentence, which is incredibly frustrating because for every sense of dictation that you do, you end up handwriting five or six words.

Perhaps the most frustrating part of it is that there is no place that it reliably works. Even in Dragon’s own notepad software, it loses track of where it is, which then prevents correction, and leaves you with the same missing letter errors seen elsewhere. I understand Dragon not being able to control how other people’s programs handle text, but not being able to make text entry work even in their own software is ridiculous.

These errors, I have no sympathy for. As I said last time, speech recognition errors are understandable because it’s a very hard task. Errors with putting text into a text box are not forgivable, and Dragon really should be working hard (and I hope they already are) to fix this, as this is not acceptable from production, $200 software.

For what it’s worth, when it gets mired in misalignments and other text errors, I found that sometimes, quitting and then reopening Dragon can be a little bit helpful. It doesn’t always fix the problem, but it’s certainly something to do if you’re partway through an email and Dragon starts to degrade on you.

Still the best out there

I still highly recommend this software to anybody who is looking to do dictation, it’s the best solution, and in many ways, the only solution. It’s still expensive, still requires a specialized microphone, and you still have the frustration stemming from stupid bugs that have nothing at all to do with speech recognition. That said, for getting a whole bunch of text down on paper very quickly, it’s still the best thing out there, and with a little bit more tender loving care from Nuance, it could be something truly astounding.

Have a question, comment, or concern about this post? Contact me!