border:1px solid green;

Simply Speak is a cutting-edge virtual scribe that saves healthcare providers time and money by streamlining the process of documenting patient visits. Simply Speak provides physicians with an automatically generated SOAP note for every patient encounter performed using our platform. Underlying this technology is an artificial intelligence platform that extracts and synthesizes useful information contained in the visit transcript. Data scientists refer to this process as Natural Language Processing (NLP). This article gives a brief, high-level overview of two examples of NLP that our data science team has built into the Simply Speak platform.

While we won’t give away any of our secrets, we do want physicians and healthcare administrators to have an idea of some of the incredible things going on under the hood in order to see the impact that Simply Speak can deliver to their practice.

As soon as a doctor starts recording a visit with Simply Speak, our platform automatically starts processing the visit, ensuring fast turnaround times. Processing starts with the automatic creation of a transcript documenting everything that was said by the doctor, the patient, and any third party. When the Simply Speak AI first encounters the visit transcript, it’s pretty messy, as you can see below:

Clinical Note-Taking Data
(Note: this is a demo visit; the transcript shown throughout this article contains no PHI.)

How Simply Speak Identifies the Different Stakeholders in the Room?

Each “item” shown on the screen above corresponds to a word spoken by the patient or the doctor. Something you might immediately notice is that speakers are only labeled “spk_0” and “spk_1”. While the automatic transcription process can tell the difference between speaker 1’s and speaker 2’s voices, it doesn’t know which one is the doctor and which is the patient.

Luckily, we can do some NLP to figure it out. In this case, we need to look at each line of text in a transcript (a line is a continuous segment of speech uttered by the same speaker) and decide, based on the information present in the line, whether the doctor or the patient is speaking in that line. We can analyze a number of features of a particular line – characteristics including how early or late in a transcript the line is spoken; patterns in the parts-of-speech of individual words in the line; whether a sentence is a question, command, or declaration; and a host of other linguistic traits. Simply Speak’s speaker detection algorithm correctly assigns the labels of doctor and patient over 99% of the time, turning a transcript like the one above into a labeled dialogue between patient and doctor:

Clinical Note-taking script

After the platform has identified which speaker is the doctor and which is the patient, it turns to complete the visit note. The first step in filling out a visit note is identifying the chief complaint – the main reason a patient is visiting the doctor.

How can we use artificial intelligence to extract the chief complaint from a transcript comprised of thousands of words?

We can look back at transcripts that humans summarized into medical notes and analyze which pieces of information humans took from the transcript and put in the chief complaint section. We can look for linguistic patterns in the transcript similar to the ones described above that predict whether a particular sentence – for example, the patient saying, “I’ve been experiencing some cold symptoms this past week” – might be helpful in determining the chief complaint. This way, we are training the AI platform by exposing it to actual past examples in which a doctor  analyzed the conversation and extracted useful information – in this case, the chief complaint – from it. When the AI is exposed to many different cases in which a transcript is reduced to a simple phrase summarizing the chief complaint, it can use statistical methods to predict which words in a new transcript should be extracted to ‘write’ the chief complaint.
These two examples provide a brief overview of the NLP methods Simply Speak uses to automate medical note taking. Implementing methods like this in our platform allows us to efficiently deliver highly accurate notes to physicians just minutes after they finish recording a visit, saving them time and cutting down on the hassle of clinical note taking.


Schedule a demo of Simply Speak. Just 30 minutes of your time will save you thousands of hours and add to your bottom line.

Daniel Vogler

Daniel is a Data Science Intern at Simply Speak. He is charged with developing and implementing natural language processing (NLP) algorithms underlying the NLP engine, contributing analytics to process improvement efforts, and supporting the development team in engineering tasks.

Daniel will graduate in May 2021 with a Bachelor’s degree from Princeton University. His academic interests at Princeton have spanned philosophy, economics, statistics, and computer science.

Contact Us

Thank you! We'll be in touch soon.

Request a Demo

Thank you! We will be in touch soon.