Too high onnxruntime version (>= 1.18.0) dependency for v5.1 ?

[JiJiJiang]

❓ Questions and Help

We have a wiki available for our users. Please make sure you have checked it out first.

First of all, thank you for your amazing open-source vad toolkit, which is mainly recommended and adopted in our speaker embedding learning toolkit WeSpeaker.

While we used the latest v5.1 version, we found that it depends on onnxruntime >= 1.18.0, which is a little bit unfriendly to some old linux systems, e.g., centos, because of its old GLIBC lib version.

I'm wondering if you guys have noticed this problem and your advice to solve it. Are you planning to support the older onnxruntime versions, e.g., v.1.16, in the next version ?

Thank you for your help!

[JiJiJiang]

BTW, could you also put v4 and v3.1 in the pip source?

[snakers4]

BTW, could you also put v4 and v3.1 in the pip source?

Since the pip-package is new (we did not make pip-packages when releasing v3.1 and v4) and the older versions are accessible via git tags, probably building several "new old" pip-package versions is possible, but we want to encourage users to use the new VAD and we really do not want to maintain the older versions of the code in the repo.

While we used the latest v5.1 version, we found that it depends on onnxruntime >= 1.18.0, which is a little bit unfriendly to some old linux systems, e.g., centos, because of its old GLIBC lib version.

I'm wondering if you guys have noticed this problem and your advice to solve it. Are you planning to support the older onnxruntime versions, e.g., v.1.16, in the next version ?

There was not much design when choosing the current minimal version. We just took the somewhat fresh version and that is it.
We will check if the current model can be run on 1.16. If so, we will just change the minimal version required.

If not, then we will have to think, how to solve this without resorting to having multiple model versions.

[JiJiJiang]

Yes, ort v1.16.3 works well with your vad model in my machine.
It is very likely that more older ort version also works. You can check v1.11 for example.

Looking forward to your update!

[snakers4]

@JiJiJiang
We checked that 1.16.1 works with the latest model.
Can you install the VAD from pip (--no-deps, --no-dependencies) and just manually install the onnx-runtime and try?

[JiJiJiang]

Yes, my temporary solution is the same as you say. I manually install onnx-runtime v1.16.3 first and then pip install silero-vad --no-deps. Everything works well.
But I think it would be better if you can lower the ort version dependency since v1.18 is not necessary.

Thank you very much for your answer.