Desktop Companions & Assistants

 

About Speech Recognition

NOTE: Microsoft's Speech Recognition is only available in ENGLISH.

Speech Recognition allows you to speak commands to your character such as launching programs or a favorite web site or even dialing the phone number of a friend or business colleague. In order to use this function you must have the Microsoft Speech Recognition Engine (currently only available in English) and a program with Speech Recognition capabilities such as CyberBuddy or Ultra Hal Speech Recognition versions and of course a microphone (a quality headset is recommended for best results). Once installed, the speech recognition engine needs to be "TRAINED" to your voice patterns and pitch etc. and also because people pronounce many words quite differently. Training for speech recognition consists of reading text shown on your screen into your microphone. As more text is read, the speech input engine learns more about your particular voice patterns, a total of 5 hours is the recommended average. It will work with minimal training so you don't have to sit and do the 5 hours or so of training all at once maybe spread it out over a period of a week or so. Remember, by choosing to use speech recognition technology, you will only get out what you put in to it. If you only do an hour of voice training and are using a cheap microphone, don't expect much more than a 65-75% accuracy in it recognizing what you say, if you do 5 or more hours of voice training and use a quality headset it's possible to achieve an accuracy level of over 90%. Results will also vary from person to person because some people speak very clear with more consistent voice patterns whereas some people speak in variable tones and may at times even mutter or mumble their words where even talking to them face to face is difficult to understand everything they are saying and would most likely achieve a much lower accuracy rate using this technology. Some people have reported receiving an 90% accuracy with just an inexpensive standard microphone and with only an hour or so of voice training, but don't assume or take for granted that you are one of those few people that has a natural clear speaking voice. You can and should by all means try starting off with using an inexpensive microphone and minimal voice training and increase your voice training and possibly upgrade to a quality headset until you achieve desirable results.

NOTE: The FREE version of CyberBuddy DOES NOT have Speech Recognition capabilities.
CyberBuddy CD Version with Speech Recognition


If you are looking for tips on how to get the most of of any speech recognition engine or to learn more about speech recognition in general, please take a look at the following links:

CyberBuddy's Speech Recognition FAQ

Zabaware's troubleshooting page

Five Tips for Better Speech Recognition

2001 A Speech Odyssey

Better but Still Not Perfect Speech Recognition

The Truth About Speech Recognition

Speak Up

Speaking to XP with SoundMax Cadenza


Microsoft Vista's built-in speech recognition is reasonably decent and may serve your purpose. XP user's can also add speech recognition for free, although not quite as robust or as good as the new Vista version, but still ok. You can download the Microsoft Speech Recognition SDK Here. Look for the "SpeechSDK51.exe" file from the list. Note: this is a 68 mb download so you'll need a high speed connection.

We have a step by step guide here for installing the XP speech recognition SDK

If you have problems installing or using it go here ... How to install and configure speech recognition in Windows XP.

Besides Microsoft's Speech recognition, below are some commercial speech recognition programs worth checking out that you can also use.

Dragon Naturally Speaking
(Research and user reports say that Dragon Naturally Speaking versions 9 and 10 offers the best the best and most accurate speech recognition presently available. The "Standard" version can be purchased for around $100 which should be enough for most users, but you can check and compare the features of other versions to see what suits your needs.)

IBM ViaVoice

RealSpeak

Philips Dictation Systems


What are the best Microphones to use for speech recognition?

First off, using a regular microphone isn't a good idea as it doesn't filter background noises properly therefore will drastically reduce the accuracy of voice recognition.

A quality headset is highly recommended. A wired (non-USB headset) is also probably the best bet as it's going to be more reasonably priced, plus quite a few of the USB headsets on the market don't work well with speech recognition, even though many claim it does on their packaging.

So which headset is the best and what type or model should I get?

There's so many headsets on the market and the models change year by year making it impossible for us to personally test or recommend specific models. However there is a company on the net that is probably the best authority on microphones, especially when it comes to voice recognition programs. They personally test various microphones and only sell models that pass their own accuracy tests. So for all intensive purposes, they are only selling the best models.

SpeechControl.com

The "Andrea ANC-700" is a good choice and is around $60. What it boils down to, is how much you can afford to spend, but the most expensive doesn't always necessarily mean better accuracy, the biggest differences between a moderately priced quality headset and the real expensive ones are technological advancements , convenience, comfort etc., for example ... if having a wireless USB model is important to you then it will be more expensive, and may or may not be more accurate than a mid ranged wired headset.

Places like Best Buy and most other computer retail chain outlets aren't the place to get advice on buying a headset specifically for voice recognition, they typically only rely on the manufacturers specs printed on the box.