Transcribing Lectures

GenAI can help with transcribing recordings. This technology has multiple uses that can save you time and expand the impact of your pre-recorded lectures for your students. We’ll dive into a few of the use cases in this section.

First, high quality recording technology is freely available on smartphones, computers, and numerous other devices today. Second, those of us that teach online courses may pre-record lectures for students to help them outside of the traditional classroom to understand a core concept, demonstrate analytical reasoning, or expand on topics covered in other materials with more recent information. Third, far more of us spend time in meetings ourselves where tracking the discussion and follow-up has typically required human intervention. GenAI provides quality solutions to automate or enhance the utility of recordings.

For example, Zoom users have the option of enabling AI Companion, which listens to the meeting and generates a reasonably reliable summary of the discussion, along with key follow-up items, which can be shared and/or edited by the meeting organizer. Microsoft Teams offers a similar service with Teams premium which is also AI-powered. OtterAI also offers a transcript generation feature for recordings.

GenAI can also unlock a superpower inherent in your lectures, which is to create a transcript that you can share with students, and that you also can use to develop supplemental materials to help student learning.

The edited transcript itself can improve the accessibility of your lecture for students that may have disabilities or are non-native speakers.[i] Transcripts also can be used by students to revisit the lecture material which can enhance student retention of the information. 

The transcript of your lecture can also be used to create a Skeletal Outline through GenAI that students can use to take notes in an orderly fashion as they listen to your lecture, helping them to identify key concepts and identify parts of the lecture they may have missed. Sweller’s research suggests that written material that complements an auditory input may help improve learning by reducing cognitive load.[ii]

That same transcript can also be used with GenAI to generate multiple choice items that you can assign for students to practice concepts covered in the lecture and confirm their understanding of the materials. A more detailed run through of this feature, focused particularly on creating importable items for D2L Brightspace, Plickers, and Kahoots!, is discussed in the next section. Practice items based on lecture content provides formative feedback, which can improve understanding and retention.[iii]

You can also use your transcript for self-evaluation of your lecture with GenAI, where GenAI can evaluate whether your lecture addresses the learning objectives you have established and offer suggestions for improvement, without subjecting yourself to a formal evaluation. A more detailed discussion of this idea is also included in another section of this book. Using self-evaluation techniques with Chat allows instructors to evaluate alignment of lectures with learning objectives in the course, enhancing instructional quality[iv] through a backward design technique.[v]

For this discussion, I’ve developed (or more accurately, co-developed with Chat) a Python script to process audio recordings through AssemblyAI. To implement this tool, you first need to create an account with AssemblyAI to obtain your own API key. AssemblyAI provides metered charges based on the length of the recording you are processing through the AI (current costs are approximately forty cents per hour of processing, with higher rates for advanced services available through their menu).

You can obtain a copy of the basic operational python code by navigating to assemblyai.com's website. If you aren't into all that, I've created subscription services on this site that permit you to upload audio recordings and obtain transcripts - no coding required!

Now, in and of itself, the raw transcript is not terribly impressive because it is just a big block of text. AssemblyAI’s transcripts are reasonably faithful to the recording and requires no pre-training on a particular person’s speech, but you may need more to make the raw transcript useful.

GenAI tools like ChatGPT can be utilized to edit the raw transcript into a more readable format. However, your prompt to Chat will need to be exact to request that the AI generate a “complete and unabridged” version of the transcript. You might try a prompt like: “Hi Chat, can you use the enclosed file to create a cleaned up and complete, literal transcript of the comments made during the recording, edited just for proper grammar and for sentence structure?”

The reason for being more specific with ChatGPT is that I’ve found it tends to summarize content by default. Specific prompting here may yield an edited but not abridged version. You may also try other GenAI tools, such as Claude.ai. In other cases, for example, when you are working with a meeting transcript, you may really just want a concise summary of the key points discussed and next steps. Your prompt’s content will lead the way. AssemblyAI also provides a summarization feature called LeMUR, where you can feed a prompt through the Python script of what should be included in the summary to generate the content you are seeking. A copy of a working Python script using LeMUR is also available for download from the book website.

Your transcript can be used for developing other materials for your course. For example, in the same interaction with Chat, you could enter an additional prompt requesting Chat “to generate ten multiple choice questions on the materials in the edited transcript to help students practice the concepts covered during the lecture.”

Such items (or a combination of true/false, matching, or other supported items – depending on your course objectives and the relative Bloom’s Taxonomy level of those objectives) could then be integrated into a course activity for students to practice and demonstrate their learning. I discuss creating items for import into a learning management system here.

If you don't want to do so much trial and error with a generative AI, I've also built subscription services into this site where you can convert your transcript into Brightspace items for import into your course.

Another technique is to generate a Skeletal Outline for students to use for taking notes while listening to your pre-recorded lecture. A prompt to Chat like “can you use the edited transcript to create a skeletal outline students can use to take notes on the key concepts covered during the lecture?” can quickly generate a document that you can edit and provide to students to use as part of the learning activity.

AI-based transcription of recordings can be a powerful learning aid for students and can help you to save time in generating related documents from meetings or other recordings. [Pane 2014; Vygotsky 1978][vi]