csukuangfj

/

speaker-embedding-models

Like 6

Hello,

We are integrating the following speaker embedding model into a video localization research project using sherpa-onnx:

Model: 3dspeaker_speech_eres2net_large_sv_zh-cn_3dspeaker_16k.onnx

SHA256:
19547e85b6c14ec44b8add4e7cb9ce353c7e995d4f1c9ffd408176ac3a2d6895

We found this exact hash in your Hugging Face repository and the sherpa-onnx GitHub release checksums.

The original ModelScope model appears to be:

• Model: iic/speech_eres2net_large_sv_zh-cn_3dspeaker_16k
• Revision: v1.0.0
• Checkpoint: eres2net_large_model.ckpt
• Checkpoint SHA256: 739b036db741e9da06448b7a533a268cdcc249d5b95d1bf46b1dec431aa5f86d

The original ModelScope model card indicates Apache License 2.0.

Could you please clarify:

1. Was this exact ONNX file converted from the ModelScope v1.0.0 checkpoint mentioned above?
2. Does this converted ONNX model retain the original Apache-2.0 license, or is a different license applicable?
3. Is the model permitted for local academic research and future commercial application use?
4. What LICENSE, NOTICE, attribution, or modification information should accompany its use or redistribution?

A reference to the original conversion record, model card, or applicable license documentation would also be very helpful.

Thank you for maintaining and sharing these models!