Policy
Voiceprint policy
Last updated: November 15, 2026
Voice recognition in LiteScribe can name speakers from a voiceprint. This page sets out what we store, where we keep it, how long we keep it and how it is deleted. It is a short version of section 8.10 of our privacy policy. If the two ever differ, the English privacy policy applies.
What a voiceprint is
A voiceprint is a set of numbers that describes how a voice sounds, not what it says. It can recognise a voice, but it cannot be used to recreate or imitate it. We treat it as biometric data and use it only to show who is speaking, for the account it was given to.
What we store
For each voice saved with consent: the voiceprint, the name it is saved under, and a record of the consent (the wording agreed to, when, and how it was given). For voices saved with the consent wording of 8 November 2026, we also keep the short recording the voiceprint was made from, so the account holder can listen to it. For "Met in person", we keep the spoken yes as proof of consent. A voice can also have a sample from each kind of device it is recorded on, for example a phone and a computer, so it is recognised on both. Each sample rests on the same consent, follows the same retention rules and is deleted with the voice.
Where it is kept
We make voiceprints on our own infrastructure, which runs on Cloudflare in the EU, and keep them in our database in Paris, France. Recordings kept with them are in our EU storage, encrypted at rest. Your phones and computers keep a copy only to recognise voices there, and a voice someone chose to keep on one device only stays on that device. Voiceprints are never sent to AssemblyAI or to any AI language model provider.
Telling speakers apart is not a voiceprint
When a recording is transcribed, its speakers are told apart and labelled, by our speech-to-text providers, by our own servers or on your device. On our servers and your devices, the voice measurements this needs are held in memory only while the recording is processed, and are never written to a database, a file or a log. We store the speaker labels, the names you give them and the words, not a voiceprint. We keep only similarity scores, which contain no voice data, to tune how closely voices must match.
How long we keep it
A voiceprint is deleted automatically at the first of these:
- You delete it, or the person it belongs to withdraws consent.
- The account that holds it is deleted.
- Your own voiceprint, or one given through "Ask for their voice": 12 months without matching a speaker in a saved recording.
- "Ask for their voice": 24 months after it was given or last renewed with "Keep my voice".
- "Met in person": 2 years after the person was last seen on your device, matched in a saved recording or met. The 12 and 24 month rules do not apply to these.
- A voice kept on one device only: 2 years after it was last heard on that device or saved, or when you sign out of the app on that device.
Recordings and consent clips kept with a voiceprint are deleted with it. Where biometric privacy laws such as the Illinois Biometric Information Privacy Act apply, this is our retention schedule and our guideline for permanently destroying biometric data.
How to delete it
If you have a LiteScribe account: go to Settings, Voice recognition, or open the person in People. You can delete your own voiceprint or anyone else's you keep.
If you gave someone your voice: use your withdraw link at any time, with no account needed, or "Who has your voice" in your own LiteScribe Settings if your account has the same email address. Someone met in person can also say "forget my voice" near the phone or computer that saved it.
Deletion takes effect at once on our main storage, reaches copies on phones and computers at their next refresh, and reaches our backups within 30 days. You can also email [email protected].
When an account is deleted
Deleting a LiteScribe account deletes every voiceprint it holds, its own and those other people gave it, with their consent records. Our daily clean-up then deletes the recordings and consent clips kept with them, within 2 days. A voiceprint of you that someone else keeps with your consent belongs to their account: it stays until you withdraw it with your withdraw link, or until one of the time limits above ends it.
What we never do
We never use voiceprints to train AI models. We never sell, rent or share them, never use them for advertising, analytics or signing in, and never use them for anything other than showing who is speaking.
Questions
Email [email protected]. We update this page whenever we change how voiceprints are stored or how long they are kept.