Hi, I've been away from all things speech recognition for a few years now and have a need again, so I thought I'd check out WSR in Windows 7.

After doing the 20 odd minutes of training, I was quite disappointed in the dictation accuracy, being used to DNS which I used to use before. I did some more training, a touch better but simply unusable for dictation, at least a word wrong every sentence, usually many.

So I reinstalled my old version of DNS (Pro 9.5), did its basic training with Kennedy's speech (I'd lost my previous training files) and it was fantastic. Totally acceptable, like I remember from before.

Despite the investment in DNS Pro, I'm still wanting to use WSR if I can get it to equal the accuracy of DNS. I just like that it's built in, the XML and caters for JScript instead of VBScript, and doesn't feel out-of-date (can't justify upgrade to DNS Pro 10 ATM).

Is it just a matter of training WSR more, or differently, or is it just not at the level of DNS yet?

Thanks

stebbo,

First, Windows Speech Recognition (WSR) is actually quite accurate when using the version in Windows 7.  The Windows Vista version isn't nearly as accurate and lacks features that are only available in the Windows 7 version.

Second, Windows Speech Recognition (WSR) necessitates longer pauses between commands, lacks the robust vocabulary of DNS, then generally requires considerably more training of individual words (correction).

Third, Windows Speech Recognition (WSR) is faster and more efficient under Windows 7 64-bit that it is under Windows 7 32-bit.

Regardless, the main problem with Windows Speech Recognition (WSR) is that it lacks the command structure (Natural Language Commands) and the additional auxiliary support provided in DNS (Acoustic and Language Model Optimizer, Vocabulary Editor, the ability to add multiple words and adapt the recognizer to your writing style, etc.).

In addition, Windows Speech Recognition (WSR) is much more sensitive and response more poorly to low-quality microphones that are not well suited for speech recognition.  It has less than optimal support for Bluetooth microphones (wireless).  It also doesn't respond well to those who have trained themselves to speak in longer more natural phrases.  It lacks a quadgram model, which DNS has, and generally tends to accommodate more dictating in short choppy phrases because it wants to display the text immediately instead of allowing the recognizer to thoroughly analyze the preceding and following words in a specific utterance (i.e., what it hears from the time you start speaking until the time that the recognizer detects that you have paused).

Just FYI, Nuance technically does not support DNS Professional 9.5 under Windows 7 (32-bit).  Also, DNS 9.5 is not compatible with Windows 7 64-bit.  For that, you would have to upgrade to DNS 10.1, as only DNS 10.1 is 64-bit compatible.  I realize that you're not going there yet but it is important that you understand this if and when you do decide to go that route.

Thus, all things considered, if you want to use Windows Speech Recognition (WSR), you need to ensure that you have a good, speech recognition certified, noise canceling microphone.  In addition, regardless of whether you're using DNS or Windows Speech Recognition (WSR), almost everyone on this forum and other forums will definitely recommend that you use a USB SoundPod such as the Buddy USB 6G or either of the Andrea USB SoundPods (half duplex and/or PureAudio – full-duplex).  The VXI full-duplex USB SoundPod is also acceptable.

You have not provided us with any hardware specifications (CPU, speed, RAM, microphone, sound card, etc.).  It would be helpful if you would do so simply because we cannot assess your experience with Windows Speech Recognition (WSR) or DNS Professional 9.5 without understanding the hardware associated with such.

Nonetheless, in terms of accuracy, Windows Speech Recognition (WSR) requires more training than what you have currently given it, as well as dependent upon the microphone/soundcard combination that you're using.  DNS is more forgiving of lower quality microphones and poor microphone/soundcard combination.  Windows Speech Recognition (WSR) is not.  Regardless, you should probably train it for another 20 minutes, and very likely get yourself another high-quality, speech recognition certified microphone coupled with the USB SoundPod as recommended above.  Doing so will probably result in much better accuracy with Windows Speech Recognition (WSR).

In addition, you might want to consider looking at VoicePower® Ultimate (www.VoiceTeach.com), which supports both DNS and Windows Speech Recognition (WSR) by providing commands features, functions and capabilities that both Dragon NaturallySpeaking and Windows Speech Recognition lack.  VoicePower is designed to work with both DNS and Windows Speech Recognition (WSR) and integrates seamlessly with both such that you have access to the same features and capabilities whether you're using DNS or Windows Speech Recognition (WSR).  In other words, you can switch back and forth simply and easily.  In addition, VoicePower® allows you to create your own personal commands without the necessity for any programming experience or knowledge of scripting.  And finally, it works identically in both Dragon NaturallySpeaking and Windows Speech Recognition (WSR).  That is, what you can do in one, you do exactly the same way in the other, and if you create your own commands in DNS, they work in Windows Speech Recognition (WSR) without any modifications or having to re-create such.  VoicePower is perfectly cross-platform compatible: Do it once, use it in both.  If you're interested and you have any questions, contact me at chuck@voiceteach.com.

The bottom line is simply this.  The software (DNS or Windows Speech Recognition (WSR)) represents only between 3% and 5% (bugs and design flaws) in terms of the problems that you can and may experience.  The remaining percentage of problems that users have with either DNS or Windows Speech Recognition (WSR) are broken down as being approximately 25% to 30% hardware and 70% user.  Even so, speech recognition is only as strong as its weakest link, and the weakest link, even excluding the user from the equation, is most often hardware (CPU, speed, RAM, microphone/soundcard).  Until the hardware is such as to be able to adequately support the software, it doesn't matter how good your dictation style, enunciation, clarity of speech, or anything related to you as a user is concerned, neither DNS nor Windows Speech Recognition are going to work properly.

Chuck Runquist
Technical Project Manager
VoiceTeach LLC
Home of VoicePower® Ultimate

Nothing in the world can take the place of persistence.  Talent, will not.  Nothing is more common than unsuccessful men with talent.  Genius, will not.  Unrewarded genius is almost a proverb.  Education, will not.  The world is full of educated derelicts.  Persistence and determination alone are omnipotent.  The slogan "Press on" has solved and always will solve the problems of the human race.  - Calvin Coolidge

-------------------------

Wow! Thanks for such an insightful response.

I'm using a Plantronics CS50 USB microphone--I need something wireless operating up to about 40 yards, ruling out the devices you mentioned. I hear what you're saying about it possibly being good enough for DNS but not WSR.

I'm only running 32 bit Windows 7, and from what you say I'm doubting I have sufficient power. Windows Experience figures of 5.1 on the e4400 2GHz processor and 5.1 on the 4GB RAM.

I think I'll muck around with WSR for a bit given your comments, possibly overclock and shut down the other 30 windows I usually have opened (!). I found the WSR scripting nicer than DNS and although DNS dictates great, I remember before always having troubles with it picking up commands quickly, even with direct training. When it does pick them up, the delay is unacceptable. WSR picks up commands instantly and correctly I found. But of course these are very quick conclusions I'm drawing.

Appreciate the trouble you went to,

stebbo

stebbo,

While I would agree with you that, to a certain extent, Windows Speech Recognition (WSR) macro scripting is a little nicer than DNS scripting because the underlying VBA compiler for DNS has not been modified since it was first introduced in DNS 6.0.  However, like I said, take a look at VoicePower® Ultimate.  VoicePower uses an XML-based Voice Scripting which is cross-platform compatible and more powerful than Windows Speech Recognition (WSR) scripting.  In VoicePower®, if you create a script using our scripting language in Windows Speech Recognition (WSR) you don't have to re-create it in DNS.  When you switch from Windows Speech Recognition (WSR) to DNS using VoicePower®, the script that you created in Windows Speech Recognition (WSR) works exactly the same in DNS with no modifications necessary.  You can even import it into DNS's Advanced Scripting without modification.  Therefore, you only have to create a script once, whether you're working in DNS or Windows Speech Recognition (WSR), and it works in both without modification.  No other application does this or is capable of such cross-platform compatibility Re: scripting.

Chuck Runquist
Technical Project Manager
VoiceTeach LLC
Home of VoicePower® Ultimate

"Aiming for the moon and missing it is better than aiming for the ditch and hitting it."   - Author Unknown

-------------------------