> ## Documentation Index
> Fetch the complete documentation index at: https://docs.medera.info/llms.txt
> Use this file to discover all available pages before exploring further.

# Diarization

> Speaker separation for conversational STT

Medera diarizes up to 4 speakers in conversational STT with sub-second labelling latency.

## How it works

* Speaker embeddings extracted per voiced window

* Online clustering with minimum-cluster-duration heuristic

* Label stability via Viterbi smoothing

* Optional enrollment for known clinicians

## Enrollment

```bash theme={"system"}
POST /api/providers/{id}/voice-enroll
Content-Type: audio/wav

```

Upload 30 s of clean clinician speech to register a speaker embedding. Enrolled speakers are labelled consistently across sessions.