Fire the transcriptionist— voice recognition works

Fire the transcriptionist— voice recognition works Until February 2004, I dictated my progress notes and had someone transcribe them. This helped me use my time efficiently, but I was paying $1,600 a month to the transcription service. Essentially, I was paying the equivalent of a mortgage payment, and it was tearing me up inside. One day, finally, I’d had enough. I decided to check out speech recognition software. I bought a copy of Dragon NaturallySpeaking Medical, version 7, and loaded it onto my IBM ThinkPad computer. Using a headset microphone that I’d found in a local office supply store, I performed the initial training with the program and then started dictating all of my patient progress notes. The software wasn’t perfect, but it was good enough for my purposes. And as I’ve learned more about how to use speech recognition in my office, it has become increasingly accurate. I now document visits in about the same amount of time that it used to take me to dictate notes for the transcriptionist. I don’t proofread every word afterwards. Rather, I check for content, and one of my staff members skims the note, looking for obvious typos and nonsense words. Because she may miss a few, I add a disclaimer at the end of each note, stating that my dictation was transcribed by computer. Here are a few nuggets of knowledge I’ve picked up along the way that can help you take maximum advantage of speech recognition. Make sure you have a good sound card. Not all computer sound cards are created equal. Some manufacturers properly shield the electrical components, but others don’t. As a result, the system can generate electrical noise that interferes with voice signals, leading to poor accuracy. If you happen to own a computer with such a sound card, there’s an easy remedy: Shell out about $25 for a USB pod and plug your mike into it to bypass your system’s sound card. How will you know if you need to do this? A reputable microphone dealer should be able to tell you after you describe your system. Get a top-rated microphone. While all microphones can detect sound input, some are much better than others at noise cancellation. If your mike picks up not only your voice, but also voices in the hallway, the sound of your air conditioner, office music, and so on, your transcription will be riddled with errors. I tried one cheap microphone after another, but none worked very well. Finally, I splurged on a $150 Sennheiser headset microphone from a company called eMicrophones (www.emicro phones.com). The Sennheiser is a version of a mike that many musicians use onstage. It’s designed to cancel nearly every other noise except for the sound of the performer’s voice or instrument. eMicrophones modifies the Sennheiser to remove a “reverb” effect that, while preferred by singers, muddies speech recognition. Headset microphones are the best kind for speech recognition, because they keep the mike at a constant distance from your mouth at all times. If you move a handheld mike away from your mouth TECHNOLOGYCONSULT The author is a family physician in San Marcos, CA. By Jeffrey Pearson, DO Fire the transcriptionist— voice recognition works ■ Speech recognition programs improve in accuracy as you use them. ■ Invest in a good microphone for better results. ■ You can never have enough speed and RAM in your computer. while speaking, this affects the sound level and results in dropped words. A faster computer is better. You can never have enough speed and RAM (random access memory). One reason why older versions of Dragon didn’t seem to work very well was that computers were a lot slower back then. Many computers sold today come supplied with 256 or 512 MB of RAM, barely enough to run the basic programs that we all have. Dragon will work on computers with that amount of memory, but it’ll be relatively slow. I’d recommend getting a system with 1 or 2 GB of RAM. If your computer didn’t come with one of those chips, buy one and put it in. Get Dragon’s medical version. Dragon comes in five flavors: Standard, Preferred, Professional, Legal, and Medical. The latter offers specialized vocabularies for cardiology, emergency medicine, family/general practice, gastroenterology, general surgery, mental health, neurology, ob/gyn, oncology, orthopedics, pathology, pediatrics, and radiology. This version sells for a bit more than $1,000 (it can be found more cheaply on the Web), but the inclusion of medical vocabularies makes it much more useful than the less-expensive programs. Speak more slowly and distinctly. I hail from New Jersey and, as a result, can speak pretty fast, so I make a conscious effort to slow down and speak more clearly when I dictate. The results with Dragon and the new Sennheiser microphone are very good, but Dragon doesn’t do well if you mumble or slur words. Of course, voices can change as the day goes on, and speech tends to become sloppier when you’re tired. But enunciation is always the key. When you first get the speech recognition program, you have to “train” it to understand the sound of your voice by reading sample medical reports provided by Dragon. You can always retrain it if your voice changes for some reason. Dragon saves the speech files from each session and continues to adjust as you use it. After a while, you should run the Acoustic Optimizer (found in the Accuracy Center of the Tools menu). This uses the saved speech files to improve the “IQ” of the program so that it works better in the future. A few final comments: I try to dictate a note after I see each patient. To speed things up, I created a template of a SOAP-style progress note. I move the cursor to the beginning of each field and dictate into it. Sometimes, when I’m running behind, I’ll have to dictate some notes later into a high-end digital recorder. But I’ve found that connecting my recorder (a Sony ICD-BMI) with the speech recognition program doesn’t produce very good results. Dragon can transcribe this dictation, but there are a lot of errors. It’s okay for jogging my memory, but not for creating an actual patient progress note. The vast majority of my notes are completed within minutes of the visit. And they’re immediately available so I can send them, if needed, with a patient to a specialist or the emergency department. Best of all, I’m saving $1,600 a month! You, too, can become a pro on speech recognition software with just a little bit of effort. If you decide to go this route, remember, use the best equipment. Also, I’d strongly recommend visiting the Web sites listed at left to find out how other doctors are using the software in their offices and what they recommend.