AUDIO · DIARIZATION
Speaker Diarization
Detect speaker-turn boundaries via audio energy analysis — processed on-device in your browser. No upload.
- 1.Add an audio file.
- 2.Adjust the sensitivity (dB) — lower values detect more sensitively.
- 3.Set the minimum segment length (seconds) to control how finely it splits.
- 4.Click Diarize, review the segment table, then download as text.
→ Output: A table of speaker segments (start, end, length) plus a downloadable text (.txt).
-60 dB (sensitive)-20 dB (less sensitive)
0.5s5.0s
⚠ Energy-based segmentation only — true speaker identification (embedding-based) is not yet supported. Segments are alternately assigned Speaker 1 / 2 at silence boundaries.
Upload an audio file and run to see speaker segments.
About this tool
Splits a recording into segments at points where the energy pattern suggests the speaker changed. It gives you rough boundaries for who spoke when in a multi-person recording. Processing finishes inside the browser.
When to use it
- —Finding turn boundaries in a meeting recording to structure minutes
- —Separating question and answer stretches in an interview
- —Getting markers to jump to one speaker's parts in a long discussion
Good to know
- —It judges by energy changes rather than voice characteristics, so it estimates where a change happened but never who spoke.
- —Overlapping speakers or similar volumes make segmentation unreliable.
- —Treat the output as reference markers; confirm the real boundaries by listening.