What are speaker labels?

Speaker labels, defined

A label is just a short marker in front of a line of speech. In a transcript it is a name and a colon. In a caption file it can be a tag the player reads as the voice of the cue. Software that works out who spoke when, without being told, does a task called diarization, and its labels are usually generic until a person renames them. See diarization.

There are two different jobs here. One is naming the speaker for the record. The other is showing that name to a viewer on screen. Those are not the same decision.

What the standards say

• The W3C accessibility pages advise identifying speakers as relevant when you transcribe.
• The WCAG 2.2 guidance on captions says they include dialogue, identify who is speaking, and carry meaningful sound effects.
• The WebVTT format has a voice span: the tag requires an annotation that is the name of the voice, which is how a caption file can carry a speaker.
• For US television, the FCC's rules count speaker identity among the non-speech information a caption must convey. Those rules apply to TV programming, not to a creator's own clips.

This is accessibility guidance, not legal advice. Whether any rule applies to you depends on where and how you publish.

Three places labels can show up

When to show a label on a clip

A clip has only seconds to explain itself, so a label is justified only when the picture cannot answer 'who is talking?'

• One speaker, on camera, whole clip: no label.
• Two people on camera, each visibly speaking when their line appears: usually no label.
• A voice from off-camera, a phone call or a guest who is not shown: label it.
• Fast back-and-forth where captions alternate: use a colour or name so lines do not blur together.
• A new speaker introduced mid-clip: give the name once, not on every line.

Worked example

Two hosts argue about pricing in a 40-second clip, and both are on camera. Labels on every line would clutter the captions, so name each person once at their first line and let the picture carry it. Now suppose one of them is only a voice on a phone line. Label every one of that person's lines, because the viewer has nothing to look at.

Common mistakes

• Leaving 'Speaker 1' and 'Speaker 2' in a published transcript.
• Mislabelling lines. A wrong name is worse than none, and automatic labelling needs review.
• Putting the label in the same style as the spoken words, so it reads as part of the sentence.
• Labelling a solo talking-head clip.

KeyMo offers 12 caption styles, including word-by-word ones such as Karaoke, and the same checks apply: watch the finished clip and ask whether a stranger could tell who is speaking. See transcript and caption.