VigOn connector — speaker-embedding model (employee-presence, voice attribution)
================================================================================
The VigOn connector downloads this file to the customer's machine as a piece of its voice artifact and
loads it ONLY when its sha256 equals the value pinned in its source (vigon-bridge
app/voice/speaker_model.py). The name carries the first 8 hex digits of that sha256: a file here is never
replaced under the same name.

voxceleb_campplus_lm-1068e4ac.onnx                    29 292 449 bytes
  sha256 1068e4ac3a76bb9c769e6816ef30bf89363f6e966f1d938210cb8ed4038f8e93
  WeSpeaker CAM++ speaker-embedding model trained on VoxCeleb (large-margin fine-tuned), unmodified,
  byte-identical to voxceleb_CAM++_LM.onnx of
  https://huggingface.co/Wespeaker/wespeaker-voxceleb-campplus-LM at commit
  c5e01c6fcffcce160861e7e79782828320192b5c.
  Attribution and licences: NOTICE (the weights are treated as CC BY 4.0,
  https://creativecommons.org/licenses/by/4.0/, attributed to the WeSpeaker authors; no changes were made)
  and LICENSE (Apache License 2.0, the WeSpeaker code).

SHA256SUMS lists every file of this folder.