Speaker Diarization Explained

Speaker diarization is a powerful AI feature that identifies different speakers in your audio and labels each section with "Speaker 1", "Speaker 2", etc.

What is Speaker Diarization?

Speaker diarization answers the question "Who spoke when?" by analyzing voice characteristics like pitch, tone, and speaking patterns to identify different speakers.

Instead of getting one continuous block of text, you get:

• **Speaker 1:** Hello, welcome to the podcast.
• **Speaker 2:** Thanks for having me!
• **Speaker 1:** Let's dive right in...

How to Enable Speaker Diarization

1. 1Upload your audio file
2. 2In the upload settings, switch to "Custom Mode"
3. 3Check the box "Identify different speakers"
4. 4Choose number of speakers: Auto-detect or specify (2-10)
5. 5Click "Start Transcription"

Auto-Detect vs Manual Speaker Count

**Auto-Detect (Recommended):**

• AI automatically figures out how many speakers
• Works best for most cases
• May occasionally over or under-identify speakers

**Manual Count (2-10 speakers):**

• You specify exactly how many speakers
• More accurate if you know the number
• Best for structured formats (interviews, panel discussions)

How Speaker Diarization Works

The AI analyzes:

• **Voice characteristics** - Pitch, tone, timbre
• **Speaking patterns** - Pace, rhythm, pauses
• **Acoustic features** - Frequency, energy

Then it groups segments spoken by the same person and assigns labels like "Speaker 1", "Speaker 2", etc.

Best Results With Speaker Diarization

• **Use individual microphones** - Each person has their own mic = much better accuracy
• **Don't talk over each other** - Overlapping speech confuses the AI
• **Have distinct voices** - Clear differences make identification easier
• **Good audio quality** - Poor audio = poor diarization
• **Avoid background noise** - Noise interferes with voice analysis

Viewing Speaker Labels

After transcription completes with speaker diarization:

1. 1Open your transcription
2. 2Go to the "Transcript" tab
3. 3You'll see speaker labels like "Speaker 1", "Speaker 2"
4. 4Each section is color-coded by speaker
5. 5Timestamps show when each speaker started talking

Editing Speaker Labels (Coming Soon)

Soon you'll be able to:

• Rename "Speaker 1" to "John" or "Host"
• Merge speakers if AI split one person into two
• Split speakers if AI grouped two people as one
• Reassign sections to different speakers

Export with Speaker Labels

When you export, speaker labels are included in all formats:

• **TXT** - Plain text with "Speaker 1:" prefix
• **DOCX** - Formatted with speaker names
• **PDF** - Professional layout with speaker identification
• **SRT** - Subtitles with speaker labels (useful for videos)

When Speaker Diarization May Struggle

• **Similar voices** - Two people with very similar voices may be confused
• **Overlapping speech** - Multiple people talking at once is hard to separate
• **Poor audio quality** - Background noise or low-quality recording
• **Many speakers** - More than 5-6 speakers becomes challenging
• **Short turns** - Very quick back-and-forth conversation

Use Cases for Speaker Diarization

• **Podcasts & Interviews** - Clearly see who said what
• **Meeting Minutes** - Attribute comments to speakers
• **Focus Groups** - Track different participant responses
• **Legal Depositions** - Identify witness vs attorney
• **Panel Discussions** - Follow multiple speakers
• **Customer Calls** - Separate agent from customer