KB1305 - Ensuring accuracy in your video captions

From VCC Disability Services:

Thank you to all instructors who are making good use of Kaltura to upload their videos.
Please be aware that there are caption editing steps that need to follow your video uploads.

Although our Kaltura system automatically generates closed captions on newly-uploaded videos, machine-generated captions are far from perfect.

Don't rely on AI hype: our Kaltura captioning engine doesn't do proper grammar, and there are many words it doesn't recognize. On average, Kaltura's machine captions may be between 80-85% accurate. That still leaves a lot of errors that need correcting.

After your captions are automatically generated, they are NOT error-free:
You should always check them and make manual corrections before your audience sees your video.

With Kaltura, we've seen both strengths and weaknesses in our machine-generated captions:


  • Canadian or American English speakers are rendered more accurately.
  • A lot of medical, legal, or technical terminology is often rendered accurately.

These strengths hint at a likely bias in how the language engine was trained by its American vendor (e.g. North American English speakers with a focus on industry/sector terminologies).


  • Non-North American speakers with noticeable accents seem to confuse the machine interpreter or lower its accuracy.
  • Poor quality audio (like a voice in a noisy background) will always produce poor results.
  • Many indigenous words and place names are unknown and poorly interpreted.

How to Improve Your Captions Accuracy

  • Record your voice using a headset or desktop microphone closer to your mouth. The "cleaner" your audio, the better your captions will be. (Don't just rely on the little microphone in your laptop or monitor. There's more background noise in your environment than you may realize. Use a headset or desktop microphone, located closer to your mouth.)
  • Manually edit your captions in Kaltura. (Fix grammar, spelling, and punctuation yourself, to achieve 100% accuracy.)
  • Help us improve Kaltura's custom dictionary to improve accuracy. eLSupport can add new words to the caption engine's dictionary when we discover words or acronyms that machine captioning has struggled with. eLSupport has added Indigenous words, acronyms (like "VCC", which used to come up as "bcc", and specialty terminology that hadn't been interpreted correctly. (Offensive or hateful words are filtered out too.)

Send eLSupport (eLSupport@vcc.ca) any words, phrases, or acronyms that seem chronically misunderstood in your Kaltura video captions, and we'll add them to the dictionary to help improve captions accuracy for everyone.

Need help learning to edit video captions in Kaltura?

To learn how to generate or edit captions in Kaltura, or to present transcripts for a captioned video, check out this CTLR online course: Using Captions and Transcripts in Kaltura.

You can always contact eLSupport@vcc.ca with any questions about video captioning.


Article ID: 8093
Thu 1/4/24 4:11 PM
Wed 2/14/24 3:39 PM