> ## Documentation Index
> Fetch the complete documentation index at: https://support.hirify.fr/llms.txt
> Use this file to discover all available pages before exploring further.

# The speakers are mixed up

> Fix speech attributed to the wrong person to get accurate interview notes back

Hirify separates the voices in an interview, but it doesn't know who they belong to until you tell it. Three symptoms can be fixed from the interview page: unnamed speakers, two people swapped, and one person split in two.

## Identify unnamed speakers

A banner appears at the top of the interview: "2 speakers aren't identified", with the note "Identifying them improves the interview notes and key takeaways."

<Steps>
  <Step title="Open the identification window">
    Click **Identify speakers** in the banner. The **Who's speaking here?** window opens.
  </Step>

  <Step title="Recognize who's speaking from what they said">
    The window shows the person's three most substantial turns, with their timestamps. **View in transcript** takes you to the right spot if the excerpt isn't enough.
  </Step>

  <Step title="Name the person">
    Enter the name in the **Speaker name** field. When participants are known, they're suggested below the field, one click away.
  </Step>

  <Step title="Set the role">
    Choose **Candidate** or **Recruiter** next to **Role**, then click **Confirm**.
  </Step>

  <Step title="Move to the next one">
    The "1 of 3" counter and the arrows take you through the remaining speakers. **Skip** leaves this one aside.
  </Step>
</Steps>

<Note>
  Two speakers can't have the same name. If you reuse a name that's already assigned, Hirify shows "This name is already assigned to another speaker".
</Note>

## Two people were swapped

This is the typical case. What the candidate said is attributed to the recruiter, and the other way around. You fix both at once.

<Steps>
  <Step title="Open the Transcript view">
    On the interview page, switch to the **Transcript** tab.
  </Step>

  <Step title="Open the menu on a turn">
    Hover over a passage attributed to the wrong person. An actions button appears to the left of the avatar.
  </Step>

  <Step title="Swap the two speakers">
    Choose **Swap with...**, then select the person this passage belongs to. All the turns of both speakers swap their attribution.
  </Step>
</Steps>

The same menu offers two other actions:

* **Rename**: fix a single name.
* **Candidate** / **Not the candidate**: a shortcut to mark or unmark the candidate.

## The same person appears twice

When a voice was split in two, for example after a microphone change or a call picked up again, the same person has two separate entries. Merge them.

<Steps>
  <Step title="Open the menu on a turn">
    Still in the **Transcript** view, hover over a passage from the entry you want to remove.
  </Step>

  <Step title="Choose the target">
    Click **Merge with...**, then the speaker to keep.
  </Step>

  <Step title="Confirm">
    The **Merge speakers** window warns you: "This can't be undone. Every segment from the source speaker will be reassigned to the target speaker, and adjacent segments will be merged." Click **Merge**.
  </Step>
</Steps>

<Warning>
  Merging two speakers can't be reversed. Check that you're merging the same person before you confirm.
</Warning>

## Carry the fix over to the interview notes

Fixing speakers doesn't rewrite interview notes that were already generated. If you just used **Identify speakers**, the banner changes to "Speakers identified", with the prompt "Regenerate the interview notes and key takeaways to reflect them." The banner has two buttons:

* **Regenerate**: reruns generation with the right names.
* **Later**: closes the banner.

After a fix made from a turn's menu (**Rename**, **Swap with...**, **Merge with...**), this banner doesn't appear. In that case, rerun generation with the **Regenerate** action, under **Interview notes**, in the interview's side panel.

## Good to know

* Identification applies to the open interview, not to others. A candidate heard in two interviews is named in each one.
* Speaker fixes don't change interview notes already sent to your ATS. Regenerate, review, then send them again yourself.
* Regenerating replaces the existing interview notes. Your manual edits to those notes are lost.
* In person, the number chosen in **Participants** when you generate the QR code helps separate the voices, but doesn't guarantee it. A very short turn or two very similar voices are still hard cases.
* If the transcript itself is empty or cut off, identification won't change anything. See the page below first.

<CardGroup cols={2}>
  <Card title="Identify speakers" href="/en/interviews/identify-speakers">
    The full procedure, outside troubleshooting.
  </Card>

  <Card title="Incomplete transcript" href="/en/help/incomplete-transcript">
    When the problem comes from the recording.
  </Card>
</CardGroup>