Dental Care Professional Assistant

Introduction

I'm looking into Speech Recognition for a while now because I think using speech can be a real game changer in a lot of use cases. This is true for all app users and not only for disabled people because there are many situations where you do not have your hands free, or the use of keyboard and mouse is just cumbersome, or you can simply talk faster than you type.
I have selected this category because the use case cannot be solved with a standalone app but can act as an addon to existing electronic health record systems. Because of the business area there might apply further restrictions and regulations for such a software.
A prototype might have been an option but wasn't feasible in the given time frame besides my other hackathon activities ;-).

My Deepgram Use-Case

The use case is from the medical area.
It is a help for dental care professionals to fill out the periodental chart. This is always done when the periodical check of your set of teeth is done.
To fill out the chart the dental care professional evaluates the teeth and records these observations

• Mobility
• Is it an Implant
• Furcation
• Gingival Margin
• Probing Depth
• Bleeding when probing
• Has Plague

This is done for the buccal and lingual parts for every single tooth. As you see, the gathered data is simple. Just an identifier for the tooth (which is a number), whether it is lingual or buccal, the observation with a boolean or number value.

No rocket science to fill out a form with this but this cannot be done by the dental care professional because he has no hands free to put it in. He must dictate it to an assistant or record the dictation and the data is put afterwards.
Both solutions double the amount of work.

Dive into Details

There are several electronic dental record systems on the market. And in one part of such a systems the teeth status is tracked.
Usually, there is a form with the representation of the teeth used to put in the values directly. Which is fine and fast if you are working in a standard desktop setting.
My proposal is to equip the dental professional with a headset (a headset filters environmental noise much better than another microphone) and a speech recognition solution based on Deepgram running on a PC with a big screen.
During the examination the professional must utter these four things for every tooth:

1. the tooth number
2. the side
3. the observation
4. the value for the observation

If these utterances are processed directly and visualized in an image of a set of teeth, the user has direct feedback that his input was correct and he can be sure to save the correct diagnosis. Being sure that everything is well understood is a problem with speech recognition and good user feedback often is key to success.

Conclusion

Why is Deepgram suitable for this? I think there are several reasons. Here are the main ones

• the latency of the recognition service is short so there should not be any disruptions because the user has to wait for the system before the next input
• a model for the use case can be defined easily and will improve recognition rate
• Deepgram service has a decent pricing

These are the main reason that let me think that this use case can be implemented with Deepgram's recognition service in a reliable and well accepted way.
But due to the forementioned restrictions this is not a task for a single developer.