Table of Contents

Enum SpeakerEmbeddingModel

Namespace
VisioForge.Core.Types.X.AI
Assembly
VisioForge.Core.dll

Selects which speaker-embedding ("voiceprint") model the diarization block is configured for. The value is a declarative hint: the ONNX file actually loaded is VisioForge.Core.Types.X.AI.SpeakerDiarizationSettings.EmbeddingModelPath, and the feature front-end auto-configures itself from that file's ONNX metadata. The enum picks the default download URL and documents the recommended clustering threshold for each model; both models are Apache-2.0.

public enum SpeakerEmbeddingModel

Fields

WeSpeakerEnglish = 0

WeSpeaker English voxceleb_resnet34_LM (Apache-2.0). Trained on VoxCeleb (English), 256-dim embedding. The default; best on English speech. Recommended ClusterThreshold ≈ 0.5 (the shipped default).

ERes2NetV2Multilingual = 1

3D-Speaker ERes2NetV2 sv_zh-cn_16k-common (Apache-2.0). Trained on a large multilingual corpus (Chinese and other languages), 192-dim embedding. Choose this for non-English or mixed-language content — it separates non-English speakers far better than the English-only model. Its voiceprints sit further apart, so it needs a larger threshold: recommended ClusterThreshold ≈ 0.8.