#Diarization — Speaker Labels

Automatically identify and label different speakers in your meetings.

#What Is Diarization?

Diarization is the process of identifying who said what in a meeting. Instead of a flat transcript, you get:

Alice: Good morning everyone
Bob: Hi Alice, thanks for setting this up
Alice: Let's start with the project update
Bob: Sounds good

Rather than:

Good morning everyone. Hi Alice, thanks for setting this up. Let's start with the project update. Sounds good.

#How Clearminutes Labels Speakers

Clearminutes uses machine learning to identify speakers based on:

• Voice characteristics: Pitch, tone, speech patterns (entirely local, no external analysis)
• Audio segmentation: Detecting speaker changes in the waveform
• Consistency: Assuming the same voice throughout = same speaker

All processing happens on your device. No audio is sent to external services.

#Enable Diarization

Diarization is enabled by default for Pro users. To verify:

1. Go to Settings → Transcription
2. Check Diarization is toggled On
3. Your next recording will have speaker labels

If diarization is off, toggle it on and re-record or regenerate transcripts.

[SCREENSHOT: Diarization toggle in Settings]

#How to Use Speaker Labels

#View Labels in Transcript

Open any meeting with diarization enabled. Each segment shows the speaker:

Alice: Good morning everyone
Bob: Hi, thanks for the invite
Alice: Let's review the agenda

#Rename Speakers

Replace "Speaker 1", "Speaker 2" with actual names:

1. Click on a speaker label in the transcript
2. Type the person's name
3. Press Enter

[SCREENSHOT: Inline speaker name editing]

Clearminutes remembers the name and applies it to other segments of the same speaker throughout the meeting.

#Auto-Detect Speaker Names (Pro)

After diarization labels speakers, Clearminutes can automatically suggest real names:

1. After recording stops, Clearminutes analyses the transcript for name mentions
2. Suggested names appear next to each "Speaker X" label
3. Review and confirm or correct each suggestion
4. Confirmed names replace "Speaker 1", "Speaker 2" throughout the transcript

This works best when participants introduce themselves or are mentioned by name during the meeting.

#Rename All Instances

Change a speaker name globally:

1. Click Edit → Rename Speaker
2. Select the speaker (or type their current label)
3. Enter new name
4. Click Apply to All

All segments from that speaker update instantly.

#Merge Speakers

If the same person was labeled as two different speakers:

1. Click Edit → Merge Speakers
2. Select speaker 1 and speaker 2
3. Choose which name to keep
4. All segments merge under one speaker

#Split Speakers

If different people were labeled as one speaker:

1. Click in the transcript where the speaker changes
2. Click Edit → Change Speaker
3. Select the new speaker name
4. Confirm

#Accuracy and Tips

Diarization accuracy depends on several factors:

#Best Conditions for Accurate Diarization

• Distinct voices: Obvious differences between speakers (tone, pitch)
• Clear audio: No background noise, wind, or distortion
• Consistent speaking patterns: Speakers talk in turns (not overlapping)
• Few speakers: 2-3 speakers is most accurate; 5+ becomes harder
• Good microphone: Professional audio input (USB headset is better than laptop mic)

#Challenging Conditions

• Similar voices: People with similar accents or vocal characteristics — diarization may confuse them
• Noisy environment: Loud background noise makes it harder to distinguish speakers
• Lots of overlap: People interrupting each other — labels may swap or merge
• Background speakers: People off-screen contributing (may be missed entirely)

#When to Disable Diarization

Consider disabling if:

• You're recording a presentation (only one speaker, unnecessary processing)
• Audio quality is very poor (may create false labels)
• You have many similar voices (confusing labels)
• You need maximum processing speed (diarization adds 10-20% processing time)

To disable:

1. Go to Settings → Transcription
2. Toggle Diarization to Off
3. Re-generate transcript (or record new meeting)

#Speaker Labels in Summaries

When diarization is enabled, summaries include speaker context:

Key Points:
- Alice: Budget increased by 20% (important update)
- Bob: We need new tools for the team (action item assigned)
- Alice: Will allocate in Q3 planning (decision made)

Without diarization, speaker context is lost:

Key Points:
- Budget increased by 20%
- Need new tools for the team
- Will allocate in Q3 planning

#Speaker Labels in Exports

When you export a meeting with speaker labels:

Markdown:

Alice: Good morning
Bob: Hi there
Alice: Shall we begin?
Bob: Yes, ready

Plain Text:

Alice: Good morning
Bob: Hi there
Alice: Shall we begin?
Bob: Yes, ready

Both formats include speaker names.

#Talk-Time Breakdown

View how much each speaker talked:

[SCREENSHOT: Pie chart showing speaker percentages]

See this in:

• Analytics dashboard (Settings → Analytics)
• Meeting details (click a meeting to see speaker summary)
• Summary (includes speaker statistics)

This helps identify:

• Dominant speakers (talking 70%+ of time)
• Quiet participants (less than 20%)
• Balanced discussions (roughly equal talk time)

#Accuracy Issues & Fixes

#Speaker labels are wrong (mixing two people)

Cause: Similar voices or overlapping speech confused the model.

Fix:

1. Manually correct labels in transcript
2. Use Rename Speaker to fix all instances
3. Lower background noise in your setup for next recording
4. Use microphones that capture distinct voices better

#Missed some speakers

Cause: Background speakers or low-volume participants not captured.

Fix:

1. Manually add missing speaker labels
2. Ensure all participants are close to microphone
3. Reduce background noise
4. Try recording with better audio equipment

#Too many false labels

Cause: Background noise, music, or echoes interpreted as speakers.

Fix:

1. Enable Noise Suppression in Settings → Recording
2. Record in a quieter environment
3. Use a directional microphone pointed at speakers
4. Disable diarization if quality is poor

#Diarization and Privacy

Diarization:

• Runs entirely on your device
• Does not identify people by name
• Does not use facial recognition
• Does not connect to external services
• Voice characteristics are analyzed locally only

You maintain full control over speaker labels — Clearminutes doesn't know or store identity information.

#Performance Impact

Diarization adds processing time:

• Recording: No impact (happens during playback)
• Transcription: +10-20% time (depends on meeting length and audio quality)
• CPU usage: Moderate increase during processing
• GPU acceleration: Speeds up diarization significantly

Example timings (30-min meeting, large-v3 model):

• Without diarization: 4-5 minutes
• With diarization: 5-6 minutes

#Related

• Advanced Analytics — View speaker talk-time breakdown
• Edit Transcripts — Manually fix speaker labels
• Generate Summaries — Summaries include speaker attribution