Imagine downloading a completely silent, instrumental video you just finished polishing in Canva, uploading it to your WordPress site, and hitting play—only to find Russian subtitles running across the bottom of the screen.
That is exactly the bizarre tech mystery John found himself dealing with. Intermittently.
If you’ve ever had a video mysteriously acquire foreign captions out of nowhere, you aren’t haunted, and your site likely hasn’t been targeted by foreign hackers. Here is the breakdown of what actually happened to John’s video and how those ghost captions got there.
The Mystery: Pure Audio vs. Ghost Text
John designed a video in Canva, used a nice instrumental backing track with zero spoken words, and verified it had no visual text or captions before exporting it as an MP4. Yet, the moment it went live on WordPress, Cyrillic text started tracking perfectly with the music.
How does text appear on a video file that didn’t have it a minute ago?
The Culprit: WordPress Auto-Transcription Plugins
The mystery isn’t actually a Canva issue; it’s a WordPress environment issue.
When you export a standard video with captions from Canva, those captions are usually “burned into” the video matrix or attached as an export layer. John’s file was clean. The captions were being generated dynamically on the WordPress side by an automated plugin or an AI media optimization tool.
Many modern WordPress sites utilize AI-driven media plugins designed to improve accessibility (ADA compliance) and SEO by automatically generating closed captions (.vtt or .srt files) the second a video hits the Media Library.
Here is exactly where the train derailed for John:
1. The “Hallucinating” AI Engine
AI transcription tools require spoken human language to function. When you feed an instrumental track (like classical music, ambient beats, or synth waves) into an AI transcriber, the algorithm experiences what engineers call a hallucination.
- The AI desperately tries to find patterns, syllables, and speech metrics inside the instruments.
- Frequency modulations in audio tracks—like a synthesized violin or a specific bassline—can occasionally mimic the phonetic frequencies of spoken Slavic languages to an automated algorithm.
2. The Default Language Bug
If a transcription plugin fails to find clear English audio, it doesn’t always just give up. Depending on how the plugin is coded, a default or fallback language might trigger if the confidence score of the audio analysis drops below a certain percentage. In John’s case, the algorithm misread the instrumental frequencies as low-quality Russian audio and tried its absolute best to translate the “music” into words.
How to Fix the Ghost Captions
If you find yourself looking at unexpected foreign captions on your own website, you can resolve the issue by taking a few quick troubleshooting steps:
- Check the Media Library Player: Go to your WordPress Dashboard, click on Media > Library, and click on the video file. Look at the right-hand details panel. Check to see if there is an attached text track or an attached caption file (
.srtor.vtt) under the video settings. If it’s there, simply delete it. - Audit Your Active Plugins: Look for plugins relating to media optimization, automated SEO, translation, or accessibility compliance (e.g., tools that auto-generate transcripts to boost search rankings). Check the plugin settings to turn off “Auto-transcribe on upload” or adjust the language detection settings.
- Mute the Source (If Instrumental): If your video is purely visual and uses background music, you can strip the audio track entirely or significantly lower its master volume inside Canva before exporting. If an AI engine scans a file with a flatlined or highly compressed audio track, it won’t try to transcribe it.
