Speaker Detection

Speaker Detection

Automatic Detection

The platform automatically detects speakers during the transcription process:

1. Voice Analysis: The system analyzes audio patterns to distinguish between different voices
2. Speaker Separation: Each unique voice is identified and labeled
3. Speech Segmentation: The system tracks when each speaker starts and stops talking
4. Confidence Scoring: Each detection includes a confidence score indicating how certain the system is about the identification

Speaker Labels

Initially, detected speakers are given temporary labels:

• Default Format: SPEAKER_1, SPEAKER_2, SPEAKER_3, etc.
• Sequential Numbering: Speakers are numbered in order of first appearance
• Consistent Labeling: The same label is used throughout a single recording
• Ready for Assignment: These labels can be assigned to named speaker profiles

Note: Speaker labels are specific to each recording. The same person may be labeled differently across different recordings until you assign them to a speaker profile.

Detection Confidence

The platform provides confidence scores for speaker detection:

• High Confidence: Clear, distinct voices with minimal background noise
• Medium Confidence: Some overlap or similar voices
• Low Confidence: Difficult audio conditions or very similar voices

You can review confidence scores when assigning speakers to help ensure accurate identification.

Multiple Speaker Handling

The platform handles various meeting scenarios:

• Two-Person Conversations: Clear speaker separation
• Group Meetings: Multiple participants with overlapping speech
• Panel Discussions: Many speakers with frequent turn-taking
• Mixed Audio Quality: Adapts to varying recording conditions