Oct 05 2015

Windows 10: Speech Recognition Performance And Accuracy


Microsoft_Windows_10_speech_recognition_thumb Windows 10: Speech Recognition Performance And Accuracy

If you’re moving from Windows 7 to Windows 10, you’ll notice a slight improvement to the speech recognition interface in the new version. In Windows 7, the speech recognition interface didn’t work at all with applications unless they had fulltext control. This is something that Nuance has done as well with Dragon NaturallySpeaking, but their customer base didn’t appreciate it as much. But the change that Microsoft made, is actually an improvement over what was available.

The speech recognition engine and Microsoft is not that bad. What’s horrendous about Microsoft speech recognition is the user interface. It lacks serious attention to detail in usability.  It has no intuitiveness to it.


Microsoft_speech_recognition_toolbar_thumb Windows 10: Speech Recognition Performance And Accuracy

So the good news is that Windows 10 includes speech recognition and that now you can dictate into applications that weren’t supported previously with Windows 7. The adjustment that Microsoft made in Windows 10 is that when you’re trying to use their speech recognition in applications that are not natively supported, with fulltext control, as you’re dictating if the application is not supported, a small dictation box pops up to allow you to enter in the information that you want and then say the word “insert”.

In my testing with Windows 10, using my WebCam as a microphone, the speech recognition was only about 30% accurate. So using speech recognition, sitting at your desktop, with the WebCam directly in front of you, don’t expect to have Windows 10 recognize your speech accurately.

But if you’re using a headset, you can expect that recognition to jump up to 80 to 90%. The recognition engine that Microsoft uses, I’m not sure what it is, but it doesn’t use the same word predictive dictionary that Nuance uses.

To be honest, I started this article with the intention of using Microsoft’s speech recognition, but it was so inaccurate and causing me to go back and edit so much, that I fell back to Dragon NaturallySpeaking [v14].

In a work environment, where employees probably are not allowed to have a copy of Dragon NaturallySpeaking, I think that Microsoft’s speech recognition can still be highly productive. Microsoft’s speech recognition can be adjusted and custom words can be added to the user library for pronunciation and proper formatting.

Microsoft_speech_recognition_speech_dictionary_thumb Windows 10: Speech Recognition Performance And Accuracy

I think Microsoft would do better for themselves if they simply offloaded the speech recognition engine to a data center for processing as opposed to allowing the PC to do the processing locally; or at least give the option.

Speech recognition in operating systems has become very competitive. Apples Siri uses the speech recognition engine from Nuance’s Dragon NaturallySpeaking. Google’s speech recognition does outstandingly well, but is not available through any operating system unless you’re using a web browser. Recently, Google started allowing users to dictate directly into Google documents.

Microsoft’s Cortana is one of the new flagship features of Windows 10, but if the speech recognition engine that they’re using doesn’t recognize the words properly, the features/commands cannot be executed correctly. With any luck, Microsoft is using the boost in Cortana’s usage to refine their speech engine.

I would still like to see all the companies that provide speech recognition offer a solution that simply uses the Internet/data center for speech recognition processing as opposed to loading bulky software on the local machines. Obviously, there would be some concerns with privacy, but there would be an option for speech recognition local processing or Internet-based processing.

Overall, Microsoft’s speech recognition works well enough to be useful, but expect some heavy editing and lots of refinements/training in the speech dictionary. The user interface for Windows 10 speech recognition is not ideal, but it is functional; it’s better than typing. Having been a user of Dragon NaturallySpeaking for years, I think that Microsoft’s speech recognition is about as accurate as version 8 of Dragon NaturallySpeaking and probably just as responsive.

Do you use speech recognition? Let me know your thoughts.


Support the site — Share this!

  • randomflux

    I normally touch type very fast but at the moment I have one hand in a cast so I’m being forced to take up dictation. I’ve been considering Dragon but in the reviews I’ve seen a lot of scary horror stories about installation nightmares, terrible support and interference with other programs. How has all that being for you?

    The worst thing about Windows speech recognition is that there’s no way to tell it whether you’re dictating or trying to issue commands and it very frequently makes the wrong decision. It’s particularly annoying in text entry fields on web forums and web pages like this, where it’s convinced the dictation is impossible and it won’t let you do it.

    • Skeeter

      I have been using Dragon for probably 6 years. I keep upgrading. Admittedly, it is very awesome and is easily one of my most used programs. I rarely write a letter, paper, or anything of substance without it. There is a learning curve, however. Basic dictation is easy to figure out but some of the advanced features, as editing with your voice, does take a little while to learn. Further, learning how to setup Advanced macros does take a little bit of time but it is well worth it as it will save you tons of time later.

      I am here today because I was curious about the performance of Nuance Dragon vs the Android voice recognition system. I recently picked up a new tablet and am shocked at how accurate and fast voice recognition has become on Android. While Android does not have the advanced editing features of nuance Naturally Speaking, it does seem to be faster and more accurate.

      • randomflux

        Hey Skeeter thanks a lot for this great information. Yes, dictation on Android has improved radically. Dictation on the iPhone and iPad is also amazingly good now. Since switching language is so easy there, I’m now using my iPhone as a dictation microphone with two Slack accounts to set up to transfer what I dictate directly from my iPhone to my PC as Slack messages.

        I bought Dragon Premium but after trying it out intensively for two days I had to return it. On my Lenovo X220 Core I7 with 16 GB of RAM the hit on the entire performance of the system was just too much to take. It also turns out that the version I was sold is not being updated at all, and to get a version within the update cycle I would have to purchase another expensive upgrade to the Professional Individual version. Since Version 13 Premium was sold to me by Nuance as the current new version this felt too much like bait and switch and I decided to cut my losses.

        • randomflux

          PS: Another great way to use dictation cross platform is to do it in Google Docs. Its new voice typing function in the Tools menu uses Google’s dictation servers directly. It’s a bit limited in many ways, but also pretty intelligent and accurate, and you can use practically any language.

          • yes; I had an article about that comparing Dragon Anywhere with Google docs/speech recognition. Dragon Anywhere is roughly $150 a year for the same services that Google Docs offers with anyone with a Google account and Internet access.

      • I’m still a big fan of Dragon NaturallySpeaking, but I have to admit that the speech recognition from Google is really good. Dragon NaturallySpeaking has voice commands and are able to format text as you’re dictating, but Google speech recognition has the ability to understand the context in which your speaking. It also has the ability to understand locations and proper names and businesses, and capitalize and punctuate the transcribed text properly.

        Over the past couple of weeks, I’ve been using Google speech recognition more and more. With very little use of the keyboard, I’m able to replace Dragon NaturallySpeaking about 90%. But this is mostly anything to do with browser-based speech recognition. I still use Dragon NaturallySpeaking for the desktop religiously.

        Microsoft has ‘ok’ speech recognition, but it’s not even close to Nuance Dragon NaturallySpeaking and far away from Google. This is one of the main reasons that I can’t use Cortana; because the speech recognition from Microsoft is so bad.

        Since Nuance made changes to Dragon NaturallySpeaking starting with version 13, that’s what pushed me to using Google speech recognition; it just works and it doesn’t require any kind of training.

        —Larry

        • Skeeter

          I have not tried, but can you use Google speech recognition for writing letters and documents and emails? Suddenly I’m impressed, as I mentioned earlier, with this while dictating on my tablet. I would love to begin writing with Google.