Glossary
De-identification
Removing or masking details that link data to a specific person, such as bleeping names or stripping metadata. A voice can remain identifying even after names are removed, which limits how far audio can be de-identified.
Techniques include redacting or bleeping spoken names, stripping file metadata, blurring faces and plates, and generalizing locations. Audio has a floor: the voice itself is biometric, so speech can rarely be truly anonymized without destroying its value. Serious datasets pair partial de-identification with consent rather than claiming anonymity.
Why it matters
Knowing what de-identification can and cannot achieve keeps the privacy claims in a data card honest.