MT Daily 

Speech Recognition, Pros and Cons

1996 discussion on SR / Back to Free Stuff

7/29/97, From: Jean Ichbiah,

Just because everybody wishes this would succeed does not mean that it will. I started working with computers in 1965. At that time the hot thing was automatic translation. Solutions would be available in five years... The research swallowed hundreds of millions of dollars and achieved essentially nothing in 30 years. The problem is too difficult.

Voice is very similar and no less difficult. The two problems are very similar in that you need to "understand" the context to do a good work. For example, it is the context that allows you to distinguish:

"urine" from "you're in"
"dilate" from "die late"
"cauterize" from "caught her eyes"
"nitrate" from "night rate"
By dealing with the phonetic level, the technology quickly reached 95%. The remaining 5% may take a century. Not 5 years. Not 10 years. Anything that requires "understanding" is bound to take that long.

At anything less than 99.5%, Voice Recognition is likely to be of little use: The time spent correcting will be more than the time you would take typing. We had this experience with OCR and found that you needed a very high percentage of accuracy to make it at all usable.

At least OCR made it easy to know what to correct. If it produced "xy69@@&&ft" your eye was attracted by the obvious typo. With Voice Recognition, their choice of seemingly correct words will make it much more difficult to spot mistakes: It will sound strange: Why do they talk about these "French eyes". You will have to listen to the tape to find out that they meant "franchise" and in that time you could have produced the report.

But it makes nice writing for jounalists, so do expect to hear more about it.

Moreover... when it comes to repetitive text, the Instant Text continuation technology can sometimes let you type faster than you can speak: you can see right on the advisory what the doctor is ABOUT to say! At one time Robert Hill thought he needed compressed speech to avoid getting bored.

Jean Ichbiah
Textware Solutions

7/97, from Holland St. John, M.D.,

I have had NaturallySpeaking for approximately two weeks. I am a physician, and for 20 years have been 7dictating my progress notes on my office patients into a tape recorder to be transcribed. This letter is being written using NaturallySpeaking. At this point, I am 99% certain that I will be continuing to do my office progress notes using NaturallySpeaking.

My throughput from starting dictation to finish, proofed, copy is now approximately 50 words per minute. This is still much slower than dictating to tape recorder, but vastly less expensive than paying professional transcriptionists. I am in a relatively low paying medical specialty, so that, even though I have to spend longer time dictating, the additional time is reasonably cost-effective. For doctors in higher paying specialties this may not be true. NaturallySpeaking is improving in speed and accuracy still very quickly. I am learning also better dictation techniques, I imagine that my dictation for a transcriptionist would be extra easy at this point.

NaturallySpeaking does as, the other poster pointed out, require much more precise dictation than most doctors are used to doing. It has taken me a fair amount of concentration to learn to dictate punctuation marks, new paragraph commands, and learning to dictate exactly what I want to have show up on the paper.

I think it will still be quite a few years before doctors in a hospital setting will be dictating directly to a computer, however I do believe that medical transcriptionists will be using this technology to more rapidly transcribe doctor's dictation than can now be done using a keyboard.

I remember that when I first started dictating reports (20 years ago) into a transcription system that it took me at least an hour for each report. Those reports typically take me (using a tape recorder) two minutes to six minutes now. Using NaturallySpeaking takes approximately two or three times that amount of time currently. I suspect that by the time two or three months goes by my slow correction techniques will be much faster, and the need to to concentrate so much on individual words will go away. That is, the extra dictation care needed with NaturallySpeaking will become a habit for me so that my speed will increase considerably.

I agree that many doctors using the current version of NaturallySpeaking would not be willing to put up with the learning curve and more precise dictation needed to use NaturallySpeaking effectively; however, the one physician that I have shown the system to has already placed his order for NaturallySpeaking with Dragon Systems.