Biometric & Voice Data Policy

Version 2026-10-25Last updated October 25, 20261,181 words

Adamic translates speech, which means we process people's voices. Some laws treat voice recordings, or information derived from them, as "biometric" data (meaning data about a person's physical or behavioural characteristics that could be used to identify them). This policy explains exactly what we do with voice data, asks for your consent before we do it, and sets out how we store and destroy it.

1. What we process, and what we don't

When you speak in a meeting and have given voice consent:

Language detection and transcription. Short stretches of your audio are sent to a speech-recognition model (OpenAI Whisper), which works out which language you are speaking and turns your words into text.
Translation. The text is translated into each listener's language and may be spoken aloud by a synthetic voice. The synthetic voice is a stock voice you choose in Settings; it is not a copy of your voice.
Speaker attribution. Captions, transcripts and action items are labelled with your name. Adamic does this by knowing which signed-in account's microphone the audio came from, not by analysing the sound of your voice.

We do not create voiceprints, meaning we do not build a mathematical model of your voice that could identify you. We do not use voice data to identify people, to infer emotions, or for advertising. Even so, because laws differ on whether raw voice audio counts as biometric information, we treat it as biometric data and apply the protections below.

2. The laws we follow

Illinois Biometric Information Privacy Act (BIPA, 740 ILCS 14). Requires a public retention and destruction policy, informed written consent before collection, and a ban on selling or profiting from biometric data.
Texas Capture or Use of Biometric Identifier Act (CUBI, Bus. & Com. Code § 503.001). Requires notice and consent before capturing a biometric identifier, and destruction within a reasonable time.
Washington (RCW 19.375). Requires notice and consent before enrolling biometric identifiers for a commercial purpose.
California (CCPA/CPRA). Treats biometric information processed to identify a person as "sensitive personal information"; you may limit its use to what is needed to provide the Service, which is all we do.
EU and UK GDPR (Article 9). Biometric data used to identify a person is special category data that requires explicit consent. We ask for explicit consent before processing voice data in any case.
Similar laws in other places, such as Colorado, Virginia, Connecticut and Texas privacy acts, are respected through the same consent and minimization measures.

The first time you join a call, Adamic shows a voice consent screen before any of your audio is sent for transcription. It explains what is processed, why, how long it is kept, and links to this policy. You choose I agree or Not now. Your choice, the policy version, the time and your browser type are logged as your written record of consent, which you can view in Settings → Privacy center.

If you choose Not now, you can still join meetings: other participants hear your original voice, but your audio is never sent to any speech-recognition or translation model, and the app shows "Translation off for you". You can turn voice processing on or off at any time in Settings.

4. Optional persistent voice profile

Settings offers a persistent voice profile option, which is off by default. Today it stores only your preference so that you are not asked again for voice consent in future sessions. It does not create or keep any voiceprint or stored audio. Minor accounts cannot enable it. If we ever add features that store voice-derived data across sessions, we will update this policy and ask for your consent again first.

4a. Speaking profile and context library (translation tuning)

If you have given voice consent and the setting Use my meeting history to improve translation (Settings → Privacy) is on, Adamic keeps a small speaking profile: average words per minute per language, median pause length, typical utterance length, how often you switch languages and which ones. These are timing statistics, smoothed over meetings — not a voiceprint, not audio, and not usable to identify you by voice. They tune when Adamic decides you have finished a sentence and which languages to expect.

Adamic also keeps a context library of names and terms (people, companies, projects, products, technical terms) said in your meetings, with the translations that were locked in, so the next meeting with the same people starts with them already correct. Only content from participants who gave voice consent is used.

Minor accounts never get a speaking profile. Turning the setting off, or revoking voice consent, immediately deletes your speaking profile and removes you from the context library (terms used only by you are deleted).

5. Retention and destruction schedule

Live audio sent for transcription is held in memory only for the seconds needed to transcribe it and is never written to disk. It is discarded immediately after processing.
Voice-derived language detection results are used only during the meeting and not stored after the session ends.
Transcripts (text) of recorded meetings follow the recording retention period: 12 months for Business rooms, 90 days for Consumer rooms.
Recordings containing your voice follow the same schedule, and include you only if you consented to recording.
In every case, biometric data is destroyed when the purpose for collecting it has been fulfilled, and never later than 3 years after your last interaction with Adamic, whichever comes first. This is the limit required by BIPA.

You can revoke voice consent in Settings → Privacy center. Processing stops immediately, including in a meeting you are already in. When you revoke, we delete stored voice-derived data linked to you: the transcript lines you spoke and any language-detection results. Recordings you consented to can be removed through a deletion request (see the Recording & Consent Policy) or by deleting your account.

We never sell, lease, trade or otherwise profit from your voice or biometric data. We disclose it only to processors that need it to provide the Service to you, under contracts that limit their use of it: OpenAI, accessed through the Emergent platform, for speech recognition, translation and speech synthesis. OpenAI does not use this API data to train its models and may keep it for a limited period, typically up to 30 days, solely for abuse monitoring. We may also disclose it if required by law or a valid court order. Other meeting participants hear your voice live, and recordings you consented to are available to room members.

8. Security

Voice data travels over encrypted connections, is processed in memory, and when recorded is stored in access-controlled object storage. We protect it with at least the same care as our other confidential information.

9. Contact

Questions or requests about voice data: privacy@adamic.io or our Data Protection Officer at dpo@adamic.io.