VOICEPRINT DELETION RIGHTS
How to Delete Your Voiceprint (And Which Tools Keep One)
A voiceprint is a biometric model of your voice — and unlike a password, you can't reset it. Courts in 2026 let plaintiffs argue that speaker labeling in meeting tools creates exactly that. This guide covers your actual deletion rights, vendor-by-vendor removal steps for Otter, Teams, and other notetakers, and the transcription path that never creates a voiceprint in the first place: local, on-device speech-to-text. VocalFuse works this way — from $5/mo.
First: does your tool even keep a voiceprint?
"Voiceprint" gets used loosely, so separate the three things transcription tools actually do. Speaker diarization splits a recording into anonymous segments ("Speaker 1", "Speaker 2"); in well-designed systems those embeddings are transient and die with the recording. Enrolled speaker profiles are explicit: you opt in, the vendor keeps a reusable model of your voice, and you can usually unenroll — Microsoft's Teams Intelliframe works this way. Persistent speaker identification is the contested middle: the tool derives voice features from every call to label "who said what," with no enrollment step and, plaintiffs allege, no consent. The February 2026 class action over Teams live transcription (Basich v. Microsoft, W.D. Wash.) claims that default diarization pipeline creates voiceprints — biometric identifiers — from every Illinois participant without the written notice, release, or published retention schedule Illinois law requires.
The quickest field test is the airplane-mode test: join a call, start transcription, then kill your Wi-Fi. If transcription keeps working, the speech pipeline runs on your machine — there is no remote server building a model of your voice, and nothing of yours is sitting in a vendor's biometric store. If transcription stops the moment the network drops, your voice is being processed remotely, and the three questions below are worth asking before your next call.
Vendor-by-vendor: who keeps voiceprints and how to get yours deleted
| Vendor | Voiceprint status | Deletion path |
|---|---|---|
| Otter.ai | Privacy policy lists "Speaker Identification / Biometric Information" among data categories; the In re Otter.AI Privacy Litigation voiceprint claims survived dismissal in August 2026. | Account-wide self-serve deletion: Profile → Account Settings → Delete account (permanent, removes conversations too; hidden while a subscription is active or a workspace owns the account). Data-rights requests go to privacy@otter.ai. |
| Microsoft Teams | Intelliframe voice enrollment is opt-in with a capture wizard; recognition features are blocked in Illinois. Basich v. Microsoft (filed Feb 5, 2026) alleges the default live-transcription diarization creates voiceprints without those safeguards; Microsoft moved to pause or trim the suit in July 2026. | If you enrolled in Intelliframe: unenroll and delete your profile from recognition settings any time. If you never enrolled, there is no user-facing voiceprint store to delete — which is part of the plaintiffs' complaint. |
| Fireflies.ai | Faces Illinois class actions (Cruz v. Fireflies.AI, reported February 2026) alleging voiceprints were captured from participants who never consented. | Submit a data-deletion request through its privacy channels and ask specifically whether speaker embeddings are created and retained; litigation makes the answer discoverable either way. |
| AI-Media | Publishes a dedicated biometric data policy — the transparency BIPA expects. | Documented: raw recordings de-identified 30 days after capture; deletion on request purges identifiable recordings within 30 days and voiceprint templates within 30 days absent a legitimate business need; templates deleted after three years of inactivity. |
| Developer speech APIs (e.g. pyannote.ai) | Voiceprint job output is transient — deleted 24 hours after job completion; the customer stores any reusable embeddings. | Deletion is in your hands by design: whatever your app persists, your app must purge. |
| VocalFuse | No voiceprints at all. Transcription runs on a local model on your Windows PC; output is text. There is no speaker biometric store and no cloud speech API. | Nothing to delete — recordings and transcripts are local files you control. Delete them like any file; they never existed anywhere else. |
What the law actually gives you
Illinois BIPA (740 ILCS 14) is the strictest US template, and it expressly names voiceprints as biometric identifiers — "biologically unique to the individual" with "no recourse" once compromised. Before collecting one, a private entity must publish a written retention and destruction schedule (§15(a)) and obtain written notice of the specific purpose and length of collection plus a written release (§15(b)); biometric data must be destroyed when the purpose ends or within the published schedule (§15(c)-(d)). The statute's teeth are private lawsuits with statutory damages of $1,000 per negligent violation and $5,000 per intentional or reckless one — which is why 2026 produced the Teams voiceprint class action, the surviving Otter claims, and Fireflies voiceprint suits. The 2024 amendment (SB 2979) limits recovery to a single violation per person per collection method, and the Seventh Circuit applied that cap retroactively in April 2026 — but the underlying right to notice, consent, and destruction stands.
In the UK and EU, voice data can qualify as special-category biometric data under GDPR, giving you erasure rights a regulator will enforce: the UK tax authority deleted over five million enrolled voiceprints after its regulator challenged the biometric database. In the US, the May 2026 nine-suit wave filed by seven Illinois broadcast journalists, podcasters, and audiobook narrators against Meta, Google, Apple, NVIDIA, Adobe, ElevenLabs, Amazon, and Microsoft (with a related action naming Samsung) seeks something stronger than damages: court-ordered destruction of voiceprints extracted from public recordings to train commercial voice models.
Not legal advice — BIPA's exact reach over diarization is being tested right now, and remedies vary by state and country. But the direction is clear: voiceprints are regulated data, and "delete my biometric data" is a request vendors increasingly have to answer in writing.
The deletion request that works
- Find the biometric policy first. Vendors covered by BIPA must publish a retention schedule. If the privacy policy says nothing about voiceprints or speaker profiles, that silence is itself the problem — say so in your request.
- Put it in writing to the privacy contact (Otter uses privacy@otter.ai; most vendors have an equivalent). Ask three things: Is a voice model or speaker embedding created from my audio? Where is it stored and for how long? When is it destroyed — and please destroy mine now.
- Cite the framework where it applies: Illinois BIPA (notice, written release, published retention schedule, destruction duty), Texas CUBI, or GDPR erasure rights if you're in the EU/UK.
- Get written confirmation of deletion. AI-Media's policy — written confirmation by email, defined windows — is what a good answer looks like.
- Remember the limits: deleting your account on one tool doesn't erase what other participants' tools captured, and voiceprints already baked into trained models may be impossible to excise (below). Prevention beats remediation.
The catch: some voiceprints can't be deleted at all
The May 2026 complaints are blunt about where deletion requests dead-end. Voiceprints extracted from public recordings to train commercial voice models aren't stored in a database that can be wiped on request — they're encoded in the model's weights. As the Google suit puts it, "the biometric data and the product are the same thing." You cannot delete an embedding that has been distributed inside a model that has already shipped. That is why the plaintiffs asked for destruction of the data and an injunction against further collection rather than a settings toggle.
Which flips the practical question. Deletion rights are real but backward-looking; the only complete protection is forward-looking — never letting a tool build a voice model of you in the first place. Prefer tools that pass the airplane-mode test, and treat "speaker labeling" in a cloud notetaker as a biometric question, not a convenience feature: ask whether it creates an embedding, where it lives, and when it dies.
The path that never creates a voiceprint
VocalFuse is built on the opposite architecture from the tools in the lawsuits. It never joins your calls — hold the pill to record from your microphone, release to paste text anywhere — and everything happens on your machine:
- Local Whisper transcription on your Windows PC — audio is not uploaded for cloud recognition
- No voiceprints, no speaker biometric store — the output is text, not voice models
- No meeting bot and no platform integration, so there is nothing for Teams' external-bot policy to block
- Passes the airplane-mode test: transcription keeps working with Wi-Fi off
- Flat pricing: $5/mo Basic, $10/mo Pro with AI summaries and Note Taking — no per-minute meter
Related reading: AI notetaker privacy, VocalFuse as an Otter AI alternative, Fireflies alternative, and Teams' notetaker-bot crackdown.
- ✓ No voiceprints
- ✓ On-device Whisper STT
- ✓ No meeting bot
- ✓ From $5/mo flat
Voiceprint deletion FAQ
Can you delete a voiceprint?
Often yes — but it depends on where it lives. If a vendor stores your voiceprint as a speaker profile or biometric template (Otter, Teams Intelliframe, enterprise speech vendors), you can request deletion, and laws like Illinois BIPA and GDPR give you leverage: vendors must publish a retention schedule and destroy data when its purpose ends. The hard limit: voiceprints already baked into a trained AI model's weights generally cannot be excised — the 2026 lawsuits against Meta, Google, and ElevenLabs argue exactly that.
How do I delete my voiceprint from Otter.ai?
Account deletion is self-serve: log in, go to Profile → Account Settings → General, and choose Delete account — Otter states account details, conversations, and associated data are permanently deleted from its servers. The option is hidden while a paid subscription is active or a workspace owns your account, and workspace members may need their admin. For data-rights requests without full account deletion, email privacy@otter.ai and ask specifically for deletion of speaker-identification biometric data.
Does Microsoft Teams keep a voiceprint of me?
Microsoft's opt-in Intelliframe feature does — but only with explicit enrollment through a voice-capture wizard, and you can unenroll and delete the profile any time. Illinois is excluded from Intelliframe recognition entirely. The February 2026 class action (Basich v. Microsoft) alleges the default live-transcription pipeline also creates voiceprints via speaker diarization — without notice, consent, or a published retention schedule. Microsoft disputes the claim; the case is ongoing. If you never enrolled, there is no Teams voiceprint store under your control to delete.
Which notetakers store voiceprints?
Any tool that identifies speakers across sessions is a candidate: Otter's privacy policy lists "Speaker Identification / Biometric Information" among its data categories, Fireflies faces two Illinois class actions over voiceprints, and Teams is in litigation over live-transcription diarization. Tools that only split audio into anonymous "Speaker 1 / Speaker 2" segments per recording — or transcribe entirely on-device — generally do not keep a persistent biometric profile. Ask any vendor directly: is a voice embedding created, where is it stored, when is it destroyed?
How do I request deletion of my biometric voice data?
In writing, to the vendor's privacy contact. Ask three things: Is a voice model or speaker embedding created from my audio? Where is it stored and for how long? When is it destroyed — and please destroy mine now. Cite Illinois BIPA (which requires written notice, a written release, and a published retention/destruction schedule for biometric identifiers including voiceprints), Texas CUBI, or GDPR erasure rights if you are in the EU/UK. Get written confirmation — vendors with good policies, like AI-Media's published biometric policy, commit to confirming deletion by email.
How do I avoid voiceprints in meetings entirely?
Use transcription that never creates one. VocalFuse runs a local Whisper model on your Windows PC: audio is never uploaded, output is text only, and there is no speaker biometric store or cloud speech API. Try the airplane-mode test on any tool: if transcription keeps working with Wi-Fi off, the speech pipeline is on your machine and no remote server is building a model of your voice.