Voice profiles and speaker recognition in workshops: what to tell your team

Written by the IdeaHone Editorial Team · · 5 min read

Colourful sound wave on a black background

You switch on speaker recognition, the transcript shows a name next to each sentence, and the first question from the room is "so it knows my voice now?". It deserves a straight answer: a voice profile touches something personal, and a team that feels it was slipped in stops speaking freely. This piece gives you the plain explanation, the data protection facts, and the words to use when you ask.

What names in the transcript are good for

Attribution is useful: follow-up is faster when the transcript says who said what, who proposed the idea and who raised the objection. Explain the mechanism before anyone records a profile, not after.

How speaker recognition works

Capture tools of this kind follow the same pattern. A person reads a few sentences once, usually 20 to 30 seconds. From that reading the tool computes a voice profile: a set of numbers that describe how the voice sounds (an embedding). The enrolment audio is not what gets kept; the numbers are.

During a session, each spoken segment is compared with the profiles of the people in the room, and the closest match gets the name. Attribution is a suggestion: the facilitator or the speaker can correct it. Similar voices, a bad microphone or a cold produce wrong matches, so treat the names as mostly right and worth a glance.

What a voice profile is not

  • Not a recording of the voice. You cannot play a profile back. There is no audio in it.
  • Not built to imitate a voice. The numbers exist to compare voices, not to generate speech, and the tool offers no way to turn them back into audio.
  • Not identification outside the sessions the person takes part in. A properly built tool compares segments only against the profiles of the people in that session. Verify this one with the vendor.

The data protection facts

A voice profile used to identify a person is biometric data, a special category of personal data under GDPR. Check the details with your data protection officer; the general shape is this:

  • It needs explicit consent, freely given, per person. A team-wide decision does not cover the individual.
  • Consent is revocable at any time, and after withdrawal the profile should be deleted immediately.
  • In an employment context consent must be truly voluntary. "No profile, no disadvantage" has to be true in practice, not just on paper.
  • The alternative always exists. Attribution by hand, or no names at all.

Recording and transcribing the session at all is a separate question; see Transcription and GDPR.

Questions to answer before you ask anyone

  1. Where is the profile stored, and is it encrypted? On the device, in your own tenant or on the vendor's servers.
  2. Who can access it? Only the people running the board, every admin, or the vendor's support team.
  3. Does it leave the organisation? Which region, which sub-processors, and whether it is used to train models.
  4. How long is it kept without use? Get a number, not "as long as needed".
  5. How does someone withdraw? A self-service button beats a request to an admin, and it must delete, not deactivate.
  6. What happens to past attributions after withdrawal? Either the transcripts keep the name as text, because the person agreed to that at the time, or they are anonymised. Decide which, and say it when you ask.
  7. Does the comparison run on the device or on a server? On-device means the audio never leaves the room for this step. Server-side means it does.

Tools like IdeaHone settle some of these by design: profiles are stored encrypted, only editors of the space receive them, profiles unused for twelve months are deleted, and each person sees in their own profile whether a facilitator recorded it.

How to say it to the team

"Speaker recognition is voluntary, and I am asking each of you individually. If you take part, the tool stores a set of numbers that describe how your voice sounds, not a recording, and nobody can play it back. You can delete it yourself at any time, and it is gone immediately. If you say no, nothing changes for you: I add the names by hand, or we work without names."

Never create a profile on behalf of someone who is not in the room, for example from an old recording. The person must be present, hear the explanation and agree themselves.

  • Enrolling everyone at the start of the workshop. Social pressure does the asking, and that is not free consent. Send the explanation a few days before and let people enrol in private.
  • Nobody knows how to withdraw. If the answer is "ask IT", most people never will. Test the path yourself first.
  • Treating attribution as truth. A wrong name in a decision log is worse than no name. Review before the transcript goes out.

Capture every idea while people talk

IdeaHone listens in your session and turns what people say into idea cards, questions and decisions, live on a shared board.

Request a demo

Is a voice profile biometric data?

Yes, when it is used to identify a person. That makes it a special category of personal data under GDPR, with explicit consent as the usual legal basis.

Can my employer require a voice profile?

In practice, no. Consent at work only counts if it is truly voluntary. There must be a real alternative, such as manual attribution or no names, with no disadvantage for choosing it. Check the details with your data protection officer and works council.

What happens to my voice profile if I leave the company?

It should be deleted with your account, or earlier if you withdraw consent. Ask before you enrol whether past attributions keep your name as text or are anonymised.

All resources