Using AI Technologies to Improve Online Course Development
by Bob Shively, Enerdynamics President and Lead Facilitator
Professionals involved in online training are experiencing rapid changes. Tools that were considered fantasies just a few years ago are now becoming essential to online course development. To demonstrate, let’s explore two examples of AI technology that we at Enerdynamics are currently leveraging for our benefit.
Narration through AI-generated voice
Our online training library includes about 50 hours of narrated courses. Up until this year, this was all developed using live narrators. This required sending a script, coaching the narrator, sending audio back for pickups when errors were discovered, and again sending scripts back for pickups when our SMEs or clients requested last-minute changes. Recent developments in AI voice generation have made this obsolete. We now use a service from WellSaid that creates voiceover that is virtually indistinguishable from narration by a person.
Using the AI-generated voice has significantly reduced our time spent implementing audio for our online courses. Our course developer can quicky generate the audio and fix any issues in minutes. And when we want to make changes or need to update a course later, there are no longer concerns as to whether our narrator is still available. We simply make the change, run it through the AI voice generator, replace the audio and we are done!
How does this work? WellSaid creates voice avatars from actual voice actors. The voice actor’s manner of speaking is captured, and the voice data is run through a proprietary deep learning model that is capable of synthesizing the actor's frequency, vibrations, breathing, pauses, and other factors that impact how we sound. The model ultimately creates a natural-sounding voice that is lightyears beyond the old computer-generated voices we all love to hate. If you haven’t heard the result, listen to one of our course demos that was created using the technology. Give it a try at this link.
Script improvement using ChatGPT
One area in online course development where we spend a lot of time is in script writing for audio. It’s not uncommon for scripts to go around multiple times among our SME, our course developer, and our editors. We now use ChatGPT to improve initial scripts to cut down on the effort required to write crisp active scripts. Here is an example of scripts modified by using ChatGPT with the command “Simplify text to be concise and change to active voice. Fix grammatical errors.”
| Original script |
Script from ChatGPT |
| The purpose of demand charges is to collect costs that are related to serving loads when they are at a maximum. These include capacity costs associated with building and maintaining the necessary generation, transmission, and distribution capacity to serve peak loads. Examples include power plants, transmission lines, substations, and distribution circuits. |
Demand charges serve the purpose of gathering costs associated with serving loads during peak periods. These costs encompass capacity expenses linked to constructing and upkeeping the required generation, transmission, and distribution capacity for peak load coverage. Examples of such capacity costs include power plants, transmission lines, substations, and distribution circuits. |
| The demand charge is based on the customers’ usage at the time that the co-op system experiences peak loads |
The demand charge is determined by the customers' usage during the period when the co-op system encounters peak loads. |
| An increasingly important additional focus for rate design is determining rates for payment for grid services provided consumer-members who own distributed energy resources. |
An additional and increasingly important aspect of rate design is determining rates for compensating consumer-members who own distributed energy resources for the grid services they provide. |
We find that using ChatGPT to test the initial language written by our SMEs results in better scripts. We don’t always accept everything ChatGPT writes, but rather have the SME compare it to the original language and then edit to add improvements. We are convinced that the result is better scripts on our storyboard as we start the course development process.
How does ChatGPT create this text? ChatGPT is called a large language model. It predicts a “reasonable continuation” of the text you give it based on an analysis of what follows similar text after studying billions of written texts available online. Or as Rodney Brooks of Robust.AI says: “What the large language models are good at is saying what an answer should sound like.” Of course, this comes with the caveat that it might give a bad answer for your situation, so while it’s a useful tool, the results require significant oversight.
AI is here to stay
Given these two examples of AI in online course development, our team at Enerdynamics is convinced AI is here to stay. We’ve added WellSaid and ChatGPT to the list of key technologies that are part of our course development process. Now we have our eyes open looking for the next innovation. We invite you to do the same.
Want to see how these technologies are creating quality online course content in the energy industry? Browse our extensive online course library here.
Back to Energy Training Today blog
