Quick question: NumPy 2.x Compatibility and ModelScope Conflicts

[dgoryeo]

您好，最近关注到了这个项目，感谢你们出色的工作！

关于依赖项我有一个小问题：目前安装 funasr-api 似乎会默认引入 NumPy 2.x。但据我了解，ModelScope（及其音频生态系统）官方目前似乎仍在使用 1.x 版本的 C-API，对吗？

请问你们在 NumPy 2.x 环境下运行此 API 时，是否观察到任何不稳定性或 C 扩展（C-extension）崩溃的情况？我担心强制升级到 2.x 可能会导致那些尚未迁移的底层 ModelScope 引擎出现问题。期待您的建议或看法。

非常感谢！

---

I just came across your repo recently—thanks for the exciting work!

I had a quick question regarding dependencies: it looks like funasr-api pulls in the NumPy 2.x stack by default. However, modelscope (and its audio ecosystem_ officially are still on the 1.x C-API, right?

Have you observed any instability or C-extension crashes when running the API on NumPy 2.x? I'm concerned that forcing the upgrade to 2.x might break the underlying ModelScope engines that haven't migrated yet. Any thoughts or nput is helpful.

Thanks

[Quantatirsk]

Thanks for raising this question, and for the kind words.

In our production deployment of funasr-api, we are currently running with the existing dependency set (including NumPy 2.x), and so far we have not observed instability or C-extension crashes related to ModelScope/FunASR.

So at least in our environment, the service has been running well in production without this issue.

That said, compatibility can still vary by platform/wheel combination. If you run into a local conflict, pinning NumPy to 1.26.x is still a reasonable temporary workaround while debugging.

We will continue to monitor this and update the repo/docs if we can reproduce any concrete incompatibility.