How to Transcribe a Meeting or Interview with Speaker Labels
Create a meeting or interview transcript with anonymous Speaker 1 and Speaker 2 labels using local browser processing.
What speaker labels actually do
Speaker diarization estimates who spoke when by comparing voice characteristics across the recording. LocalScribe assigns anonymous labels such as Speaker 1 and Speaker 2. It does not recognize names, verify identities, or determine a person’s role.
The same person should usually keep one label, but this is an estimate. Similar voices, overlapping speech, background noise, and very short replies can split or merge speakers incorrectly.
How to create the transcript
- Prepare the recording. Use a file in which all participants can be heard. Obtain permission where required.
- Open Speaker Diarization. Add the local audio or video file.
- Select the language. Choose the known spoken language or use automatic detection.
- Enable Speaker labels. This adds separate local speaker-analysis models and extra processing time.
- Start transcription. The transcript is generated first, followed by speaker analysis and timestamp alignment.
- Review every label. Compare speaker changes with the recording and correct the editable text before using it as a record.
- Export the result. Download TXT, SRT, or VTT. Speaker prefixes are included when labels are available.
Improve the source recording
- Place the microphone where every participant is audible.
- Reduce music, room echo, keyboard noise, and side conversations.
- Ask participants not to speak over one another when possible.
- Avoid aggressive noise removal that makes voices sound distorted.
- For remote calls, use the clearest recording you are authorized to access.
These steps cannot guarantee correct labels, but clean, separated speech gives the analysis more useful evidence.
Turn anonymous labels into names carefully
Listen to the opening turns and map labels to names only when you know the participants and are authorized to identify them. Check later sections because a diarization error can cause labels to switch. Do not treat voice clustering as identity verification.
For formal minutes, legal records, research interviews, or publication, a person should review the wording and attribution against the source recording.
Privacy and processing time
Transcription and optional speaker analysis run on your device. Enabling speaker labels downloads additional model data and performs extra local processing, so it takes longer and uses more memory than transcription alone. Failure of the optional speaker step should not discard a successfully generated transcript.