100% dictation accuracy with speech recognition?

Mark Beans, Director, Channel Programs & Service DeliveryCustomer Care, Developer, Healthcare, Partner Leave a Comment

When using speech recognition, you may be accurate 97% or 98% of the time, can you get to 100% accuracy? Well, nobody speaks perfectly all the time. But if you did speak flawlessly with no mistakes or stumbles you might get to 100% accuracy!

Think about some of these true life scenarios though, I happened to assist in onboarding a dentist to our speech platform. He was doing well, had great accuracy, was very pleased with our speech recognition product – and had substantially improved his daily productivity in working with his patients. Then one day, I happened to be on an email trail in which he said that while everything had been working perfectly, but at 4 PM the day before his accuracy went down considerably.

About the Author

Mark Beans, Director, Channel Programs & Service Delivery

Share this Post

I jumped into the SayIt Admin Console to review his recent dictations and the problem was obvious. Listening to his dictations, there was an overbearing whirring noise in the foreground of that dictation and every dictation from then on. I emailed the doctor and told him that he needed to either turn off the grinder or move his dictation station. He was astounded and effusive in his praise. “That’s amazing! How did you know that was the problem? It’s working great again!”

One of the advantages of cloud based speech recognition is that we can help to diagnose problems that users are having because we can listen to their dictations on the platform.

What’s the take away? Practice a little common sense ownership of your own dictation accuracy. If your accuracy isn’t what you think it should be, listen to a few of your own dictations!

Think about it, every speech recognition system on the market provides you the ability to listen to your own dictations, so why wouldn’t you listen to yourself occasionally? Especially if your accuracy doesn’t seem as good as you think it should be.

Supposedly Abraham Lincoln once said, “Give me six hours to chop down a tree and I will spend the first four sharpening the axe.” Well, in speech recognition you won’t need four hours to prepare and improve your efforts, but you might need to take 30 seconds here and there to listen to yourself speak.

Spending 30 seconds to listen to your own dictation occasionally might save you thirty to sixty minutes a day or more!

Here are just a few of the things that I have heard in the middle of what otherwise should have been very accurate dictations by users:

  • Very loud hydraulic doors in the ER
  • Coworkers laughing and cutting up
  • A very loud receptionist answering the phone right beside the dictator

Each time these background noises were apparent, the speaker’s accuracy dropped substantially.

If you have loud noises going on right beside your dictation station, then maybe you need to move your dictation station. Maybe that’s not the best location to stand and dictate.

I heard a clinician dictating one time with very loud coworkers standing right beside him. Every time the clinician stopped to think and collect his thoughts, the speech of his coworkers was returned in his pauses with perfect accuracy. The funny thing is that I don’t think he realized that their speech was being reproduced. Maybe he thought there was a “ghost in the machine”? He was probably just focused on his work but it must have been very frustrating to him. It would have been to me.

Step away! Or ask your buddies to take it outside. Or perhaps you need to invest in a decent quality microphone with noise cancelling capabilities! A three to four-hundred-dollar microphone may seem expensive, but how much is your time worth? A good quality microphone might last five to ten years and there are extremely high quality, noise cancelling headsets available for less than $40.

I have been asked to listen to some dictation samples with poor accuracy only to find that the microphone quality was so poor a human couldn’t even understand what was being said, much less a speech recognition engine. Sometimes, even high quality microphones fail. In one of those situations, I asked the clinician where he got the microphone he was using. The doctor told me that he borrowed it from his son’s Nintendo as he was leaving the house that morning (true story). Hmm… and you wonder why your accuracy is poor? This microphone wasn’t designed for use with speech recognition.

Another doctor that I worked with was a great guy, very pleasant, but ah, very ah, disfluent in his ah, speech. I worked with him over a period of several days coaching him to think about what he wanted to say before dictating. Unfortunately, he struggled for several weeks. Then one day, miraculously it seemed, his accuracy jumped from the low 80’s to the mid 90’s! I called him and asked him what made the difference. He told me that he changed jobs. The poor guy was experiencing so much stress in the company that he was working for that he was only getting about 3 hours of sleep a night. When he changed employers, he found a more relaxed atmosphere in his new job and suddenly his disfluency disappeared literally overnight. When his speech disfluency disappeared, his speech recognition accuracy jumped by more than 10%. That’s a pretty big improvement!

So what’s the point in all this? You might need to take a little ownership for your own success with speech recognition. Can you get your dictation accuracy to 100%? Well, obviously, that depends on how good a speaker you are. Many customers get 100% dictation accuracy on many or even most of their dictations; but in a nutshell, the most powerful tool for speech recognition accuracy is you!

Leave a Reply