Blog

How to Fix Speaker Labels in a Transcript

Author: Hushscript | Published: October 6, 2026 | Last reviewed: October 6, 2026

To fix speaker labels in a transcript, do the work in the editor: rename a label once to change every line it owns, reassign a single turn to a different speaker, and split or merge turns where the boundary is wrong. You do not need to transcribe the file again. Automatic labels are a first draft, and a few minutes of review makes them reliable.

Rename Speaker A to a real name

Automatic labels are neutral. The transcript shows Speaker A and Speaker B because the system groups voices; it does not know who they are. Once you recognize the people, open the speaker names dialog and type a name for each label. The name replaces the label on every line that speaker owns, in the editor and in exports.

You can also use a role instead of a name. Interviewer and Participant work well when names should stay out of the file. If Insights exist, Hushscript can suggest speaker names in a reviewable block. Nothing is applied until you accept it. The speaker identification page explains what the labels do and do not claim.

Move a wrong turn to the right speaker

Sometimes one line carries the wrong label while the rest of the speaker is fine. Change the speaker on that turn only. Pick the right speaker from the list, or choose a new speaker if the line belongs to a voice the transcript missed.

Work through the transcript with the audio playing. Check the first line after each pause and each point where the topic changes. Those are the places where a label is most likely to flip.

Split a turn that holds two speakers

When two people talk in quick succession, the engine can place both in one turn. Split the turn at the point where the second voice starts, then assign the second part to the right speaker. Keep the split points tight so each part holds one voice.

Merge turns that belong together

The opposite problem is one speaker’s sentence cut into two turns, often by a pause or a cough. Merge the turn with the next one to join them. This only works cleanly when both turns have the same speaker, so fix the label first.

One person can also appear under two labels, for example Speaker A and Speaker C. There is no dedicated command that merges two speaker labels. Reassign the turns that carry the extra label to the correct speaker. When none remain, that label is gone from the transcript.

Every edit can be undone and redone, and the whole transcript can be reverted to the original if a pass goes wrong.

Why speaker labels come out wrong

Knowing the cause tells you where to look. The guide to how speaker diarization works goes deeper. These are the usual culprits:

• Overlapping speech. When two people talk at once, one voice often dominates. Expect missing or merged turns around interruptions.
• Similar voices. Two speakers with close pitch and tone can collapse into one label.
• Phone and video-call audio. Compression and a changing connection alter how one person sounds, which can split them into two labels.
• Short turns. A one-word answer such as “yes” gives the system little voice to work with, so it can attach the turn to a neighboring speaker.
• Many speakers. The more people, the more boundaries to get right. A panel of eight needs more review than a two-person interview.

Review is the answer in all of these. Automatic labels are not a verified record.

What to set before you transcribe

Some problems are cheaper to prevent than to repair.

Turn speaker detection on for any recording with more than one voice. It adds no minutes. For a solo recording, turn it off.

If each person was recorded on a separate channel, use multichannel handling. It keeps the channels apart, and each channel is billed. It still produces transcript labels, not isolated voice tracks.

Good recording habits help more than any setting. Give each person their own microphone when you can. Keep the mic close and the room quiet. Ask guests not to talk over each other, and have everyone say their name at the start so you can match voices to labels fast.

The proper format of a speaker label

There is no single required format. The common convention for written transcripts is a name or role at the start of each turn, followed by a colon. Choose one style, such as Maria: or Interviewer:, and keep it for the whole transcript.

Exports follow that pattern. In TXT, DOCX and the subtitle formats, each turn or cue starts with the speaker name and a colon. Markdown puts the name in bold above each turn. CSV and TSV have a speaker column. JSON keeps the raw label on each utterance, with the saved names alongside. For TXT, Markdown, HTML, RTF, ODT and PDF you can switch the speaker labels off if you want plain text.

Subtitles have one more option. The WebVTT specification defines a voice span for naming the speaker of a cue. Hushscript’s VTT export writes the name as plain text at the start of each cue instead, so any player shows it as ordinary caption text.

Rename the speakers before you export. The export uses the names as they stand at that moment.

Edit the labels or transcribe again

Edit in almost every case. Corrections are free, immediate and keep your other edits.

Transcribe again only when a setting was wrong from the start: speaker detection was off, the wrong language was chosen, or separate channels were not handled as channels. A new transcription is a new job billed for the full audio length, and the new transcript starts without your manual edits. Fix the setting in the options, then run the file once more.

If you plan to generate Insights, finish the labels first. Insights are tied to the transcript version, and generating them again later uses minutes.