Introduction
It started with what looked like social media content — short clips filmed on a tablet, full of filters, music, and on-screen gestures. The videos were recovered during a routine safeguarding investigation. On first pass, they appeared harmless. Nothing explicit. Nothing shareworthy.
But in the background of several clips, quiet voices could be heard beneath the music. Muffled, hard to decipher and definitely not part of the child’s narration.
Due to the volume of material and the masking effects from overlays and audio tracks, we couldn’t manually review it all with confidence. We suspected grooming, but we had no clear starting point — and no audio specialists on site.
We weren’t even looking at the background audio at first — it was just noise. But S21 made the invisible, visible. That changed everything.
We dropped all 90+ clips into S21 Transcriber. No audio stripping, no reformatting. The tool recognised the media and began processing, video and audio, immediately.
Despite the music overlays and sound effects, the transcription engine picked out hidden dialogue. Background voices emerged from the noise. Adult speakers were recorded asking the child for personal details, arranging times to call and referencing chats that weren’t part of the clips.
The system flagged key phrases like “don’t tell”, “our secret”, and “wait until they’re gone”. Each line was time-stamped, speaker-separated and linked to exact moments in the videos. We could click straight from transcript to playback — no guesswork, no scrubbing.
What was most concerning, but most useful, was the discovery of multiple adult voices. That insight shifted our approach from a safeguarding flag to a wider exploitation investigation.