campplus-speaker / SOURCE.md
lwh2017's picture
Add SOURCE.md
fac3b17 verified
|
Raw History Blame Contribute Delete
1.74 kB

Provenance

Field Value
Upstream model 3D-Speaker CAM++ speech_campplus_sv_zh-cn_16k-common
Upstream home https://www.modelscope.cn/models/iic/speech_campplus_sv_zh-cn_16k-common
License Apache-2.0
ONNX export sherpa-onnx release speaker-recongition-models, file 3dspeaker_speech_campplus_sv_zh-cn_16k-common.onnx
Export URL https://github.com/k2-fsa/sherpa-onnx/releases/download/speaker-recongition-models/3dspeaker_speech_campplus_sv_zh-cn_16k-common.onnx
Artifact assets/campplus_zh.onnx, 28281138 bytes
Artifact sha256 f682b514c05d947ee3fa91cd6ec6c5c7543479a128373fa29b1faedccd21fd11
Modifications None to the graph. This bundle adds inference_manifest.json, assets/adapter.json, README.md and this file.

The ONNX file carries the upstream identity in its own metadata, which is what ties this artifact to the model card above:

framework=3d-speaker  language=Chinese  sample_rate=16000  output_dim=192
normalize_samples=1   feature_normalize_type=global-mean
url=https://www.modelscope.cn/models/iic/speech_campplus_sv_zh-cn_16k-common/summary
comment=This model is from iic/speech_campplus_sv_zh-cn_16k-common

The copy packaged here was downloaded from the export URL on 2026-10-22. It was not possible to re-download it from the packaging host to confirm byte identity with the release asset (GitHub is unreachable from there), so the sha256 above is the authoritative identity of what this repository contains; compare it against the release asset before trusting the two to be the same bytes.

Repackage with:

ros2 run voice_asr_service package_campplus_bundle \
  --bundle-root models/campplus-speaker --workspace .