Voice recognition software now feasible

Tech Talk. By Tom Savel, MD, amednews contributor. Jan. 29, 2001.

TECH TALK

A column about putting technology into practice

SEE ARCHIVES

So, what's the scoop with voice recognition software? Why hasn't it caught on like wildfire in clinical practices? Let me give you a brief overview of some of the key issues involved.

Voice recognition technology lets you speak into your computer's microphone and have "correct" words generated in, for example, Microsoft Word. Over the years, the technology has improved drastically. Thanks to the development of what's known as "continuous speech recognition technology," gone are the days when you had to pause after each word.

Links

Currently available voice recognition software packages allow you not only to dictate into your computer, but to actually control your computer with your voice. For example, you can say, "Open Microsoft Excel," and then "Open file 'patientlist.doc.' " You can even navigate through the Internet with this software. Clearly, this feature has a lot of potential.

Thanks to another technological advance called natural language recognition, you have the ability during your dictation to say, for example, "Boldface the last sentence," or even, "Go back one sentence and delete the first word."

Available software and hardware

There are several different packages on the market. Some of the most popular programs are made by IBM and Lernout and Hauspie. Those who are familiar with the software by Dragon Systems may know it is now a part of Lernout and Hauspie. The software packages range in price from less than $100 to more than $1,000. I have found (as did CNET.com) that the IBM product had the best combination of price and performance.

Any software package comes with a headset microphone. But given the work flow in your office, running over to a PC and slipping on a headset may not be ideal.

One alternative is the digital recorder. By speaking into this recorder the same way you spoke into your portable cassette recorders in the past, you then can transfer your voice to the PC later.

As to hardware, start with a very solid PC. Aside from what the minimum requirements may say on the outside of their packaging, voice recognition applications are hungry for both processor power and hard-drive space. For example, ViaVoice, IBM's voice recognition product, when completely installed, uses more than 500 megabytes of space.

My advice is, whatever you do, if you have more than one hard drive, don't install the software on your main Windows drive (usually the C drive). In general, use at least a Pentium II or equivalent microprocessor. Although 64 megabytes of RAM (memory) will work, more will afford you better performance.

Voice issues

What are some of the key issues related to voice recognition technology and its applications in clinical practice?

First, your learning curve.

I know it's very hard to find the time, but you need to set aside at least two to three hours to get everything working (installing the software, testing the hardware and reading the manual). Remember, you will also need to take the time to decide where you will be doing the recording.

In addition to learning how to have the PC correctly recognize your voice is the extremely important process of correcting mistakes. Understand that the program will make mistakes, and it's truly a matter of your ability to quickly correct each mistake. Some choose to use their voice to correct the mistake, while others find it faster to use a combination of their voice, the mouse and the keyboard. Over time, you'll find the ideal combination.

You need to decide how to integrate this new dictation method into your daily clinic work flow. Clearly, if you prefer to dictate after each patient visit, you can't afford to be fumbling with new software during a busy clinic.

Changes to your daily routine are never easy. I'd recommend some practice dictations after hours with the help of your office staff. During the practice session, be sure to make some mistakes and then correct them.

Try to do your dictations in a relatively quiet area, or at least an area with a consistent level of noise. If there is one thing software packages like, it's consistency. Consistency helps the program understand the difference between your voice and noise.

Now, we come to your software's learning curve.

As I said earlier, consistency is key for essentially trouble-free voice recognition. Remember that just as you need to learn how the software works, the software needs to learn how you work -- your specific voice.

If you read into the microphone for 30 minutes, the software will greatly increase its accuracy. In addition to the 30 minutes of recording, the software will take quite a few minutes to process your voice. It doesn't matter whether you have an accent, as long as you are consistent with it. Don't be surprised if your accuracy goes down when you have a cold -- your voice will sound different to the software.

Next comes your clinic's learning curve.

Your clinic staff needs to adjust to your new dictation method. This may involve not only a change in the placement of medical charts but also an awareness of when you dictate, when not to interrupt you and when to keep the noise level low.

Continuous voice recognition software does work. Without question, there is a definite learning curve for all involved parties, and that's why it hasn't caught on like wildfire.

For those who choose to take the time, they should find the voice recognition experience quite rewarding. Given the need for ever-improving documentation, as well as the ever-tightening budget, voice recognition software is definitely worth a try.

Dr. Savel is president and co-founder of Cedar Cove Technologies, a clinical software development company. He can be reached by email (tsavel@cedarcovetech.com)