Alkd/campplus-zh-cn-common-200k-onnx

CAMPPlus zh-cn-common-200k (ONNX)

Speaker embedding model for diarization, exported from ModelScope damo/speech_campplus_sv_zh-cn_16k-common.

Specs

• Architecture: CAM++ (Context-Aware Masking)
• Training data: CN-Celeb + CN Common (~200k speakers)
• Embedding dim: 192
• Input: feats — [batch, 80, time] (80-dim FBANK, 16 kHz, 10ms hop)
• Output: embs — [batch, 192]

Compatible with

pyannote-rs EmbeddingExtractor (same feats/embs tensor names as wespeaker_en_voxceleb_CAM++.onnx).