piaopiao

Results 5 comments of piaopiao

原因是因为我们当前构建的镜像是使用FROM nvcr.io/nvidia/pytorch:23.09-py3,NV构建的镜像是没有torchaudio的,我并不期望在镜像中重新使用pip3 install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121 来安装,这样会违背了使用NGC的pytorch容器的初衷(避免自己安装torch和Nvidia环境),我们目前没有找到更适合的镜像信息,因此看一下FunASR是否可以在没有torchaudio的情况下只使用librosa 我看了一下当前项目源码中使用torchaudio的,基本上异常处理都会包含一次使用librosa处理的方案,除了torchaudio中使用kaldi的没有,因此这里我直接将kaldi代码从torchaudio中抽出来,这样可以最小化修改代码的前提条件下达成我的目的 我对kaldi中的代码没有更多的研究,单纯的简单处理了一下,我希望你们可以有更好的方案来帮忙进行处理这个问题,或者有其他更好的办法来完成我的想法和需求,非常感谢 这是我的简单修改 https://github.com/Honst211/FunASR/commit/ea8f43437428478ec445b6332e1363a9fc4f60ca#diff-46d81aff6f65ae24bab6475a62c0a6796b31790f7a8e0dfc7f9078c4ab0b77f4

> How you install it? If both funasr and whisperx need to use the latest, this is the installation order of my current Dockerfile. First install the latest funasr, modelscope,...

I read what you discussed. The main problem at present is that the docker image does not have **torchaudio**, but we need **torchaudio** to run **modelscope**. I read the source...

This warehouse was modified by me so that I can use FunASR without **torchaudio**. The test showed that there is no problem. Check to see if you need to replace...

No, I asked the official about the PR of this requirement, and they said that the code of this warehouse will no longer be maintained. . . , let me...