Original URL: http://www.theregister.co.uk/2012/09/10/review_nuance_dragon_dns_12_voice_recognition_software_review/

Dragon Naturally Speaking Premium 12 voice recognition software review

Give your PC a hearing aid

By Simon Williams

Posted in Software, 10th September 2012 11:00 GMT

Speech recognition has been a technology coming of age for an age. It got a shot in the arm recently with the launch of the iPhone 4S, where the S stands for Siri, the speech recognition company Apple bought. Siri may be trendy, but the most mature technology is on the PC and comes from the company Nuance bought.

Nuance Dragon NaturallySpeaking Premium 12

Voice training on Nuance's Dragon NaturallySpeaking Premium 12

Dragon NaturallySpeaking, now in version 12 (DNS 12), is just about the only dictation and command speech program on the PC worth talking about. Although the minimum spec suggests it would run on a netbook, Nuance recommends a 2.2GHz, dual-core CPU.

Furthermore, if your PC starts to struggle, the new version enables you to turn off batches of recognition commands you’re not using, to improve recognition speed. For instance, if Start menu or desktop voice commands won’t be used, DNS 12 can be set not to check for them.

Nuance Dragon NaturallySpeaking Premium 12

All that speech power in one little bar

Install the software and the DragonBar appears on the desktop and can be floated or latched to top or bottom of the screen. The supplied headset (you can opt for a Bluetooth version) plugs into mic and headphone sockets and includes stereo earpads – previous ones were one ear only. Pre-training isn’t needed, though it can improve recognition accuracy.

The program is designed to enable spoken commands and dictation – you can even get a DNS package with a Philips voice recorder thrown in. However, the range of commands in Office isn’t new to version 12 although there have been some improvements here.

Nuance Dragon NaturallySpeaking Premium 12

Functional headgear: intelligibility is key to voice recognition with dodgy audio, rather than actual speech, being a major cause of errors

As well as the obvious formatting commands, such as New paragraph, Select sentence, Caps on/off, DNS can access any of the Ribbon commands in Office apps and now requires the word ‘Click’ to precede them, to help distinguish between commands and dictation of words such as Home, Insert, Review and the like. So, starting a sentence with Insert, won’t be confused with saying Click Insert, to switch to the Insert tab.

Mail outs

A new feature Nuance is talking up is that many of DNS’s dictation commands can now be used in on-line e-mail services, too, so you benefit from this level of control in the composition fields of Hotmail and Gmail. It can also be used for most Windows commands and is ideal for those who work on their own or who speak loudly on mobiles in trains.

Nuance Dragon NaturallySpeaking Premium 12

On-line e-mail services benefit from a side panel of command prompts on the right
Click for a larger image

The other four key new features in DNS 12 include a 20 percent improvement in out-of-the-box accuracy, the ability to use an Android phone as a wireless microphone, an interactive tutorial and longer and better chosen lists of alternatives, when the program mis-recognises.

Version 11 of Dragon NaturallySpeaking already claimed 99 percent accuracy, meaning on average one error every 100 words, so you can now expect one error every 120 words. Under test, I saw rather more, though in dictating a review like this – with a lot of numbers and technical terms – it still did remarkably well. The overall accuracy of the program depends heavily on the quality of the sound stream it hears. It will, for example, give better accuracy with a wired, noise-cancelling headset, than from a DECT or Bluetooth mic, which in turn will be better than the microphone in a typical mobile.

Nuance Dragon NaturallySpeaking Premium 12 Nuance Dragon NaturallySpeaking Premium 12

Android remote access now supported

I mention the mic in a mobile as DNS 12 adds Android support to the iPhone support it already provided, so you can use a phone as your input source over a Wi-Fi link. When I tried this, it worked, but it isn’t as convenient nor as accurate as a dedicated headset. You can also use a phone, media player or portable recorder to record notes and then transcribe them from the recording. Again, the better the mic quality, the better the text file the program produces.

Another aspect that Nuance endeavours to address in DNS 12 is user preferences for how certain words or phrases should appear. The Smart Format Rules feature attempts to learn these quirks as you go, but for technical writers, it's not as clever as I'd like it to be.

Nuance Dragon NaturallySpeaking Premium 12

Limited facility to make global formatting changes

Most UK titles require numbers below 10 to be spelt out with those above to appear as numerals and DNS 12 gets this, so we're doing fine so far. However, most also require units of measure to follow directly after a number, with no space between. So, I write five kilograms (below 10), but 15kg, not 15 kg, the latter being the default in DNS. Alas, you have to change individually every abbreviated unit of measure to remove the preceding space. There’s no global facility for this.

Vocal opposition

It would be very helpful to be able to instruct the program to change the default format for all unit abbreviations, rather than having to set each one individually, which is still the case in DNS 12. Indeed, there's plenty of room for Smart Format Rules to get smarter. If these bugbears seems a little niche then, for mainstream users and those coming to DNS for the first time, the new interactive tutorial will prove useful. However, the requirement to repeatedly click on the program’s microphone icon before every exercise and on retries, quickly becomes tedious.

Nuance Dragon NaturallySpeaking Premium 12

Most Nuance DNS Premium 12 commands work on words rather than phrases

Now if you're wondering about Apple's alternatives, there’s one key difference between Dragon NaturallySpeaking and Siri, the speech recognition in the iPhone 4S and in OSX Mountain Lion. DNS converts speech as you talk, so, if necessary, you can say a short phrase or a single word and have it recognised very quickly on the computer on which it’s installed.

Siri, on the other hand, does its recognition on servers in the Cloud – devices send information in larger chunks, to keep the overall recognition rate up. This means you have to dictate largely blind and it can make correction more fiddly. It also means you have to have a Cloud connection to use it and that recognition speed depends in part on the connection speed.

Also, voice recognition on the iPhone 4S is primarily intended for commands, rather than dictation. AI is used to interpret a wide range of queries which aren’t required in DNS. DNS is designed to operate Windows, major applications like Office and to take dictation.

The range of dictation controls in Siri, even on Mountain Lion Macs, is smaller than in DNS and correction of mis-recognised text relies on simpler commands. The two technologies are aimed primarily at different tasks and each is well tailored. Personally, I’d rather have the near-instant recognition provided by DNS 12, though this isn’t always possible on lower spec devices.

Verdict

Although it’s the only choice in town for PC speech recognition, Dragon NaturallySpeaking hasn’t sat on its laurels. Version 12 really does seem a bit more accurate, a bit swifter and offers useful extensions, like better control of on-line email. There are still niggles, like number handling, but it proves to be a lot quicker than typing. ®

More Voice Recognition Product Reviews

Google
Nexus 7
Evernote Apple
Mac OS X 10.8
Mountain Lion
Vlingo Apple
iPhone 4S