Semantic twins in voice recordings and images
How Nakato finds the details that identify someone in a call recording and a photo, replaces them with a twin, and keeps what the task needs.
- Length
- 0:59
- Published
Read the transcript
You can now protect identities in voice recordings and images. Take this repair request. Here’s the raw audio. [Caller] Hi, it’s Sarah Jones at 14 Mill Lane. My tenant number is 4176. The handle’s come off again. We find the caller’s name, street address and tenant number in the recording. We replace all three to create a semantic twin. We also change how her voice sounds so she can’t be recognised. Listen again. [Changed voice] Hi, it’s Emma Reed at 26 Park Road. My tenant number is 8325. The handle’s come off again.
Those same details are printed above the photo in her repair report. We find them in the image and use the same replacements you just heard, matching the original typeface and spacing. Then we blur her face. The repair team can still see exactly where the handle came off. Put your data to work. The real details stay with you.
More explainer videos
1:35 ·
Watch a semantic twin at work
This video explains how Nakato swaps identifying details for a twin that keeps the context a task needs. An AI tool can reason properly while the real details stay with you.