Review: Dragon NaturallySpeaking lets you take your hands off the keyboard

When Gene Roddenberry first came up with Star Trek in 1964, having the ship's computer understand natural speech was an obvious choice—after all, the series was about the far future. Teleporting isn't quite here yet, but speech recognition very much is.  Dragon NaturallySpeaking, now in version 12 ($200, buy only), is one of the veteran products in this category. Though it's pricey, it remains the best speech-to-text program for Windows.

$200 is a good-sized chunk of change, especially for a program that doesn't have a trial version. What's more, productively using speech recognition requires more than just software: You need a decent microphone, a quiet environment where nobody minds you talking to yourself, and a different state of mind than when composing text using a keyboard. The good news is that if you're just curious about speech recognition and want to try it out, it's probably built into your system: Starting with Windows Vista, every version of Windows includes a speech recognition feature. This also means Dragon faces stiff competition: $200 with no trial, vs. a free option that's already installed on your computer.

To compare the two, I took a paragraph from the Wikipedia entry about Alice in Wonderland and tried dictating it using both products. It wasn't a very scientific experiment, but I did use the same text, computer, microphone, and environment. Dragon's rendition was noticeably better, but both results were far from perfect, mainly because the paragraph contained several proper names. Still, the results indicate that Windows speech recognition is definitely usable, especially if you just want to evaluate if speech recognition makes your work any faster or easier.

The added value Dragon offers over Windows speech recognition comes in two forms: Quality of recognition, and extra features. Disregarding the single-paragraph test, I found Dragon's recognition excellent, especially for general prose. I was able to dictate a lengthy email without having to correct many errors, using my natural voice and without having to enunciate or speak any differently than I would to another person. In other words, Dragon's core functionality feels mature and field-tested, and it does work.

It's hard to quantify, but it does feel a bit faster and more accurate than the previous version, NaturallySpeaking 11.5. Even when Dragon misunderstood my speech, the correction alternatives it offered were usually accurate and often included what I was actually trying to say (and when they didn't, I could teach Dragon new words).

The other part of the added value equation, extra features, is less even. Some are fantastic, like the ability to feed Dragon NaturallySpeaking Premium with an audio file and have it transcribe the text. This lets me record audio using my smartphone when I'm away the computer or out of the house, and delivers excellent results. Other features aren't as polished: There's no obvious way to switch between two USB microphones (like a webcam and a headset) without re-calibrating the microphone.

Dragon now supports a Remote Microphone app for Android and iOS, letting you use your smartphone as a wireless microphone for live dictation. To configure the Android app, you can scan a QR code—but in my case, the QR code shown by Dragon contained incorrect information about my computer's IP address, making it impossible for the Android app to connect with the desktop app (Dragon picked a virtual network adaptor rather than the real one). Fortunately, you can also configure the Android app manually, which I did.

The Android app worked well with Dragon as long as my phone's display was switched on, but the moment I switched it off, it stopped working (even though the phone was still connected to my network even with the display off).

A better new feature for Dragon 12 is Webmail integration: Dragon 12 comes with browser extensions that make it possible to work with Gmail and Hotmail. I tested it with Gmail and it worked well, letting me pick a recipient, specify the subject, and dictate the email. I also liked that it doesn't let you actually send the email using only your voice, which is great for preventing accidental sending.

Also new in Dragon 12 is a "getting started" tutorial providing a brief overview of Dragon's capabilities and commands. It's much shorter than the Windows speech recognition tutorial, but it's a welcome addition for bringing newcomers up to speed and showing how exciting speech recognition can be.

NaturallySpeaking includes numerous features other than dictation. You can format text, launch applications, search the Web, and more. Many of these features are indispensable for users with accessibility needs, and it's great to see them built into the main version of Dragon rather than into a specialized product.

You don't have to use these advanced features to benefit from Dragon, though: If speech recognition is a good fit for your environment and workflow, Dragon's excellent dictation feature alone makes it a good investment.