ERes2Net-base speaker embedding (3D-Speaker, zh-cn), ONNX
An unmodified copy of 3dspeaker_speech_eres2net_base_sv_zh-cn_3dspeaker_16k.onnx from the
sherpa-onnx speaker-recongition-models release,
mirrored here so that mac-toolbox can download it from a pinned revision.
- Original model:
damo/speech_eres2net_base_sv_zh-cn_3dspeaker_16k(3D-Speaker, Alibaba DAMO Academy), Apache-2.0. - Trained on the 3D-Speaker dataset (10,000 speakers, multiple devices, distances and dialects).
- Input: 16 kHz mono audio. Output: a 512-dimensional speaker embedding.
- ONNX export: k2-fsa/sherpa-onnx.
- SHA-256:
1a331345f04805badbb495c775a6ddffcdd1a732567d5ec8b3d5749e3c7a5e4b(matches the sherpa-onnx releasechecksum.txt).
mac-toolbox 的聲紋比對用這個模型;檔案沒有改過,授權照原模型(Apache-2.0)。