The lesson video she used to record four times
Grace Tan teaches financial accounting at a polytechnic in Singapore. Her module runs as a flipped class. Every Sunday night her students get a short video that walks through one worksheet, and on Wednesday morning they come to class having watched it, or at least that's the idea. Class time is for the questions the video raised.
I'm David, and I work on VoxStudio. Grace is made up, but her Sunday is built from what teachers have told me about making lesson videos. Explaining the worksheet takes her twelve minutes. Getting twelve clean minutes on video used to take most of an evening.
Week 6, take four
Last term she recorded with the webcam app that came with her laptop and a separate screen recorder, then lined the two up by hand. In week 6 she was nine minutes into a walkthrough of adjusting entries when she said "debit the prepaid rent account" and meant credit. She noticed it while she was still talking. She had no easy way to cut out one sentence, so she started again from the top.
The second take had a phone notification in the corner of the screen. The third had her neighbour's renovation drilling under it. The fourth was fine. Then came the captions. Two students in the class are hard of hearing, and the school asks for captions on recorded lessons. She typed them herself in a free subtitle tool, line by line, until a quarter to one.
There was one more thing. About a third of the cohort are international students from China. Several had told her, politely, that the video was fine but the terms went past too quickly. "Accrued expense" and "contra account" are hard enough in your first language.
One window, one take
This term she records in the VoxStudio Mac app. She opens Create, then Record and transcribe, and picks Window instead of the whole display. She chooses the spreadsheet window, so the recording shows only that. Other apps and notifications are not recorded. She turns on the camera and picks her USB microphone. The worksheet has no sound of its own, so she leaves system audio off.
The creator workflow guide suggests a short test first, and she does one every week. Ten seconds, replay, check that her face, the screen and her voice are all there. It caught a muted microphone once, which would have cost her the whole evening.
Then she teaches. When she trips over a sentence now, she stops, takes a breath, and says it again properly. She doesn't start over. Fixing it is a later job, and a short one.
The transcript arrives with the recording
When she stops, a preview opens. She trims the first few seconds, where she is reaching for the mouse, names the session "FA week 6 adjusting entries", and goes on to Processing options. Local processing is the default. The recording is transcribed on her Mac, and VoxStudio Cloud is only used if she ticks it. She adds a few hotwords for terms the model might not expect, such as accrual, prepaid and contra. She also turns on Segment subtitles so the transcript comes back as caption-length cues.
The session opens with a transcript beside the player. She reads it once against the audio, mostly checking numbers, because an accounting lesson is full of them and "fifteen" and "fifty" are easy to mishear. She fixes two.
Cutting out the wrong word
Next she chooses Create Clip. The video editor opens with her screen and her camera on separate video tracks, plus the audio and the subtitle cues. That only works because the recording held both sources, which is why she records them together in VoxStudio rather than importing a flattened video.
She types into the editor's AI Chat: "Keep the spreadsheet as the main view. Put my camera small in the bottom right with rounded corners. Don't cover the captions." She checks the preview, nudges the camera up a little by hand so it clears column H, and moves on.
Then the slip. In the word-level cutting guide, the razor shows the subtitle split into the words before and after the cut point as you move along the timeline. She slides it until the marker sits just before "debit the prepaid rent", clicks, then slides to just before her corrected sentence and clicks again. The cut goes through the video, the audio and the subtitle at the same point. She deletes the piece in the middle, closes the gap, and plays across the seam from a second before to a second after. It sounds like she said it right the first time. She reads the captions on both sides of the cut too.
Before exporting she uses timeline search, the magnifier in the timeline, and types "prepaid rent". The edit-by-transcript page explains that exact matches come from the subtitles on the timeline. She clicks through each result and listens for which side of the entry she named, to make sure she hasn't made the same mistake anywhere else. She hasn't, this week.
The captions she used to type until one in the morning came with the transcript. She styles them in the Captions panel, larger than the default because students watch on their phones on the bus. The captions guide is clear that captions rendered into a video and an exported SRT file are separate outputs. Her school's video platform takes the rendered version, so that is what she exports. Then she opens the file and watches the first and last minute before uploading it.
A transcript the class can read slowly
For the international students she does one more thing, and it takes about five minutes. The handout comes from the session, not from the edited video, so she first corrects the slipped line in the transcript. Then she clicks the translate icon beside the Transcript tab and picks Simplified Chinese. The Mac translation guide says plainly that speech recognition runs on the Mac, but translation uses the AI provider set up in Settings, so the transcript text goes to that service. Her lesson is a worked example about a made-up company, and nothing in it is about her students. She is fine with that. She wouldn't do the same with a recording of a student consultation.
She switches between Original and the translation and reads through the accounting terms. She reads Chinese well enough to spot one that came out as an everyday word instead of the accounting term, and checks the right one with a colleague who did her first degree in Shanghai. Then she exports the transcript with the Bilingual variant as a TXT file and posts it next to the video. It isn't a subtitle track. It's something students can read at their own pace with the worksheet open.
In week 7, one of those students came to class with the bilingual transcript printed out and three terms circled. That was the conversation Grace had wanted to have all along.
What it takes
VoxStudio for Mac needs an Apple silicon Mac (M1 or later) and macOS 15 or later, and you download the local speech models once before transcribing. It's free for 14 days. After that, you can keep recording for as long as you need, but the free version transcribes only the first 15 minutes of each recording. The creator workflow tutorial walks through the whole path with a short practice recording.
If you make lesson videos, try it on next week's. Record it in one take, mistakes and all, and see how long the fixing actually takes.



Comments
Post a Comment