Does Voice Recognition Software Work?

Voice recognition software has been on the market for years. It is a software program that lets you talk into a microphone and translates your words into text. The early versions didn’t work so well, but the new generation software seems to be outstanding because of the statistical and modeling systems they use. It is complicated, but think about it—a program that has a 60,000 word vocabulary must look through trillions of possibilities when words are strung together.

You have probably encountered a voice recognition system already. Many companies use a recorded system that you speak to when choosing from the menu options. Sometimes it will tell you that it didn’t understand your response, so you will need to speak more clearly and directly into the receiver. If it still can’t understand you after a few attempts, it will automatically transfer you to a live person. Also, several cell phones have voice activated dialing, so when you enter someone’s phone number into your phone book, your phone will prompt you to say the person’s name as well. Then when you want to call someone, you can say, “Call Bill” instead of dialing the phone.

When you buy a package, such as Dragon NaturallySpeaking or iListen (for Mac users), you will generally get a microphone and the materials you’ll need to “train” the software to recognize your voice and speech patterns. Once you’ve done the initial 15 minutes of training, which consists of reading some text to the program and correcting any mistakes it makes, the accuracy is 90-95%. Once you’ve used the program a lot and corrected the errors, the accuracy comes close to 100%.

Voice recognition software is great for those who are disabled, people who transcribe documents, and businesspeople who dictate their notes, memos, to-do lists, and correspondence. There are several different packages that are available, and many are customized according to needs. The cheaper programs will translate slowly and not be as accurate, but may suit your needs just fine.

How does all of this work? Just as the alphabet has 26 letters that we recognize, voice recognition programs have 44 “phonemes” that it recognizes. These phonemes are small bits of speech that are analyzed and identified and paired up with text equivalents to come up with the spoken word.

Despite all the great benefits of voice recognition, there are still some flaws in the programs. First, the program needs to be able to hear you, so any noise that competes with your voice will interfere. You should use your program in a quiet room and make sure you speak directly into your microphone. This also means that the software can’t be used in a meeting where several people are talking over each other or interrupting each other.

The software needs a lot of computer power, and you’ll get a slow response time if your computer doesn’t have enough processor speed. You’ll need a lot of hard drive disk space too because the vocabularies are extensive.

Homonyms also cause some interesting problems. Sometimes the software won’t know the difference between “to”, “two”, and “too”, or “petal” and “pedal”, or any other words that sound the same. The statistical models should help with the context, but it doesn’t mean that the right word will always be chosen. Of course you will need to proofread carefully after you finish your dictation. Running the document through a spellchecker isn’t going to be enough, as some words will be spelled correctly but be the incorrect word. There is a context checker that will try to make sure the right word is being used, but it isn’t foolproof.

If you’ve ever wondered whether or not voice recognition software is right for you, give it a try. You’ll be surprised at the time you can save and will be pleasantly surprised with the results if you take the time to properly train your system.