# Replicate published Whisper Diarization example

Source: [thomasmol/whisper-diarization on Replicate](https://replicate.com/thomasmol/whisper-diarization), the published default example titled **AI should be open-sourced**.

Published example prediction: `9hzpcnz0knrmr0cyp6zr17v390`. Model version: `744c4f2bffae674f82e79ed46f9cc54796d4903b3e20e68353e05a93eb10a55c`. Retrieved from the read-only Replicate model API on 2026-10-23.

The website uses one continuous excerpt at 24.12–52.34 seconds (28.22 seconds). Five complete original model segments and their word entries, speaker IDs and probabilities are retained. Only start/end timestamps are shifted by 24.12 seconds. No text or speaker correction, voice synthesis, stitching, denoising or new inference was performed. TXT/SRT/VTT/normalized JSON use the website's existing exporters. MP4 and poster are a generated waveform visualization of that same excerpt, not interview footage.

This is a **provider-published demonstration**, not a website-generated result or an accuracy benchmark. Display names in interactive rename examples are illustrative. Speaker labels and estimated word times may contain mistakes. The original published example supplied an exact count of two speakers and the vocabulary prompt `LLama, AI, Meta.`; the current website's upload flow does not expose a vocabulary prompt. No affiliation or endorsement by the interview participants, model author or Replicate is implied.

Source recording and interview rights remain with their respective holders. The provider's model license does not itself license the source interview; no CC BY license, permission grant, or ownership of the interview is claimed here. The excerpt is included to demonstrate and inspect the linked provider example at the site owner's request. The separately retained AMI test clip has its own documented CC BY 4.0 provenance.