Speaker Diarization
What is Speaker Diarization?
Libraryminds Speaker Diarization is a built-in capability. Speaker diarization splits a transcript by voice, so each paragraph is attributed to Speaker 1, Speaker 2 and so on. You turn it on when you upload a recording, and you can rename the speakers afterwards. It is available on Plus, Pro and Business plans.
Speaker diarization for meetings — How teams use speaker labels in meeting transcripts.
Speaker diarization, explained — What diarization is and how it works.
- ✓ Paragraphs are attributed to separate speakers in multi-voice recordings
- ✓ Turned on with a switch when you upload a file
- ✓ Rename a speaker label to a real name, up to 60 characters
- ✓ Works for meetings, interviews and panels
- ✓ Labels carry into the transcript view
- ✓ Needs Plus, Pro or Business
Available on the Plus plan and above.
How it works
- When you upload a recording, turn on the speaker diarization switch. It is off by default and needs a Plus or higher plan.
- Libraryminds asks the transcription service to separate the voices as it transcribes.
- The finished transcript attributes paragraphs to speaker labels.
- Select a label and rename it, for example from Speaker 1 to the person's name. The new name replaces the label in the transcript. If you use the same name in several transcripts, a badge shows how many transcripts it appears in.
Good to know
- You choose diarization when you submit the recording. A transcript made without it does not gain speaker labels afterwards.
- Labels are a best guess from the sound of the voices. The service can merge two similar voices or split one person into two, so check a few paragraphs before you rely on them.
- Very long recordings are processed in parts. The labels come from each part, so check that the same label means the same person all the way through.
- It does not know who anyone is. You supply the names by renaming labels. A name you reuse is matched by its text, not by the sound of the voice.
- Overlapping speech and cross-talk are the hardest cases and the likeliest to be mislabelled.
- Free plans do not include speaker diarization. The switch is shown but needs Plus or higher.
- The public page for a shared transcript shows times and text, not speaker labels.
When diarization helps, and when to skip it
| Recording | Diarization | Why |
|---|
| A meeting with several people | Turn on | Attributing decisions and actions needs to know who said them |
| A two-person interview | Turn on | Separates the interviewer from the guest |
| A lecture or monologue | Skip | One speaker, so labels add nothing |
| A noisy room or a call with crosstalk | Turn on, then check | Results are less reliable |
| A recording you will quote from | Turn on, then check each quote | Never attribute a quote without listening |
Get cleaner speaker labels
- Record each person on a clear line where you can. Separate microphones give the cleanest separation.
- Ask people to speak one at a time and to avoid talking over each other.
- Have the host say each person's name at the start. It is easy to rename labels when you can hear who is who.
- Keep recordings in one piece where possible. Splitting a meeting into files restarts the labels in each file.
- Spot-check the first few turns and a stretch near the end, then fix the label names.
After the transcript: put the names in
Renaming a label is the step that turns a transcript into a record. Replace each generic label with the person's name once, and every paragraph by that voice changes at once. Keep the names consistent, for example always first name and surname, so that searching for a person finds every line.
Use Cases
Attribute decisions and actions in a meeting
Turn on diarization for a team call, rename the speakers, and read who committed to what.
Separate interviewer and guest
Use labels to pull out the guest's answers and quote them accurately.
Follow a panel discussion
See how views were distributed across several speakers, then search for one speaker's contribution.
Frequently Asked Questions
Which plans include speaker diarization?
Speaker diarization is available on Plus and higher plans. It is not included on the Free plan.
Do I have to turn it on every time?
Yes. It is a switch on each upload and is off by default.
Can I add speaker labels to an existing transcript?
No. Diarization is chosen when the recording is submitted. To get labels, upload the recording again with the switch on.
Can I rename the speakers?
Yes. Rename a label to a real name of up to 60 characters, and it changes throughout the transcript.
How accurate are the labels?
They are a best guess from the sound of the voices. Similar voices can be merged and one voice can be split, so check before you attribute anything.
Does it work on long recordings?
Yes, but long recordings are processed in parts, so check that a label means the same person throughout.
← Transcription & Input features · All features