Speakers & Diarization
Use speaker diarization in Kalima to separate voices, assign and reuse names, keep labels consistent in transcripts and exports, and understand limitations.
Kalima automatically tells your speakers apart and labels each part of the transcript, so a multi-person recording reads as a clean, attributed conversation instead of one wall of text. You can replace the generic labels with real names, and those names follow your transcript everywhere, into the live view, shared sessions, and every export.
How speaker detection works
Speaker diarization is the process of distinguishing and labeling the different voices in your audio. Kalima does this automatically while it transcribes, so there is nothing to switch on.
- As different people speak, Kalima groups their words and assigns each voice a label: Speaker 1, Speaker 2, Speaker 3, and so on.
- Each speaker gets a distinct color so you can scan who said what at a glance. The color palette has eight colors and repeats for additional speakers when a recording has more than eight.
- Speaker numbers stay unique across the session, including when you stop and start several recordings. Automatic diarization does not recognize a person's identity across takes, so the same person may receive a different number in a later recording.
Diarization is automatic, but it is a best-effort estimate. Very short interjections, similar voices, or overlapping speech can occasionally be attributed to the wrong person. Use Check voices after recording to review and correct the grouping.
Check voices after recording
For a more careful review, open Check speakers and choose Check voices. Kalima listens to the recording again on your device, groups passages by voice, and shows samples you can play before applying changes.
The voice check runs entirely on your device. No audio leaves it. It needs at least two passages with speech and takes about 40 seconds per hour of recording.
Naming your speakers
Replacing "Speaker 2" with a real name turns a raw transcript into a professional, readable record. Names appear in the segment headers, on the speaker badges, in the live view, and in every export you generate.
Once you rename a speaker, every segment attributed to that voice updates at once. To undo a name and return a speaker to its default "Speaker N" label, open the same badge and use the clear option.
Reuse names you have already entered
When you open the rename popover, Kalima shows the names you have already used elsewhere in the session as quick-pick chips. This keeps names consistent across multiple recordings in the same session and saves you retyping, which helps when one person speaks across several takes of a long meeting.
Only the session owner can rename speakers. If you are viewing a session shared with you, you will see whatever names the owner has assigned, but the badges will not be editable. To change a name, ask the owner, or open the session from the account that owns it.
Where speaker names appear
A name you assign is not just a label in the editor. It shows on the speaker badges and in the segment headers as you read and refine, and Kalima carries it through the rest of the experience too.
Names in your exports
Speaker names are pulled fresh each time you open the Export dialog, so your exports always reflect the most recent names you assigned. Any speaker you have not given a custom name to keeps its localized "Speaker N" label by design. If an export still shows "Speaker 1" after you renamed someone, simply close and reopen the Export dialog so it picks up your latest names. See Exporting transcripts for the full list of formats.
Use cases
Diarization and naming shine anywhere more than one person speaks:
- Interviews. Label the interviewer and each subject for a clean, quotable record.
- Panel discussions and meetings. Produce attributed minutes that show exactly who decided or committed to what.
- Classes and lectures. Separate the instructor from student questions.
- Multilingual conversations. Speaker labels pair naturally with the per-segment language badges so you can follow both who spoke and in which language.
Limitations to keep in mind
- Speaker renaming is owner-only. Viewers of a shared session cannot edit names.
- Names are limited to 50 characters.
- The color palette repeats after eight speakers, so the ninth speaker shares a color with the first.
- Diarization is automatic and may occasionally misattribute very short or overlapping speech.
Frequently asked questions
Click the speaker badge (it shows a small pencil icon) in any segment header, type a name up to 50 characters, and confirm. Only the session owner can do this. The name then appears everywhere: badges, segment headers, the live view, and all exports.
No. Speaker separation is automatic and always on. You just record, and Kalima labels the voices as Speaker 1, Speaker 2, and so on. You can rename them at any time, during or after the session.
The speaker color palette has eight colors and repeats for additional speakers. If your recording has more than eight distinct voices, the ninth onward reuse colors from the start of the palette. The numbers stay unique even when colors repeat.
Not necessarily. Numbers remain unique across the session, but automatic diarization does not recognize identity across separate takes. Use Check voices and reuse the same name to bring the transcript together.
No. Only the session owner can rename speakers. Viewers see the names you have assigned but cannot change them. If a name is wrong, the owner needs to update it.
Export names are loaded when the Export dialog opens, so reopen the dialog to pick up your latest renames. Also, any speaker you have not given a custom name to will keep its "Speaker N" label by design. Assign a name to any speaker still showing the default.