Issues running Torchaudio using the Nvidia Container

Robotics & Edge Computing Jetson Systems Jetson Xavier NX
, ,
1

I’ve been working with the Nvidia Container used in the Getting Started with AI course. I’m trying to experiment with torchaudio, but when I import it. I get this error:

Torchaudio Import Warning
Torchaudio Import Warning1345×53 2.39 KB

When I’m checking on the torchaudio utils file, I get an error specifically from the local import of the audio backends
Backend Utils.py importError

The problem seems to happen when attempting a relative import inside of the container, as it happens inside of the sox_io_backend.py file as well:

.common import error
.common import error728×196 4.5 KB

My torch version is 1.9.0, and my torchaudio version is 0.9.0a0+33b2469

The jetpack version I found from the command apt-cache show nvidia-jetpack is:

Jetpack Version
Jetpack Version975×347 12.5 KB

I am using a Jetson Xavier NX from Seeedstudio. The link to the product is here: Jetson SUB Mini PC-Blue with powerful Jetson Xavier NX module, aluminium frame enclosure, 128GB SSD, WIFI, antennas and pre-installed JetPack system.(Ready to use) - Seeed Studio . Any help on this would be appreciated

2

Hi @spolk, I would recommend upgrading your JetPack to JetPack 4.6.1 (L4T R32.7.1) and use one of these containers that have the SOX backend built for torchaudio:

  • nvcr.io/nvidia/l4t-pytorch:r32.7.1-pth1.9-py3
  • nvcr.io/nvidia/l4t-pytorch:r32.7.1-pth1.10-py3
  • nvcr.io/nvidia/l4t-ml:r32.7.1-py3

Previous versions of these containers (which the DLI container is based off) did not include the fix (although you could rebuild them yourself from the latest dockerfiles at https://github.com/dusty-nv/jetson-containers if you don’t want to upgrade your JetPack)

If you want to use JupyterLab, then I’d recommend using the l4t-ml container which has JupyterLab server already installed.

3

Hello,
Updating to Jetpack 4.6.1 fixed all Issues I was having with torchaudio. Thank you for the help!