Why can I not install soundfile in a Tensorman container?

[veirs]

I can use tensorman run --gpu etc. to create a container. But, when I install libraries, one library does not appear to install. Why?

python3 -m pip install matplotlib numpy scipy soundfile # installs all these dependencies fine when accessed from outside of the container in a terminal window.

But, inside the tensorman container, the matplotlib, numpy, scipy import fine but the soundfile dependency does not.
import soundfile # returns
OSError:sndfile library not found

The soundfile library reads and writes audio files and I wonder if there may be some subtle permissions issue?

Help!!!

[mmstick]

When you start a container from an image, the container's environment only exists in memory. You have to save the container to a new image to retain any changes you made to it.

[veirs]

The problem seems to be that yes, I can make changes to the image and yes, I can save the changes and then, later, run the new image. But, in contrast to all the other dependencies that I installed in the new image, soundfile (https://github.com/bastibe/python-soundfile) 'installs' it seems but then is not recognized when import soundfile is executed in python. All these same procedures work properly when run outside of the container.
I will attach a txt file with snippets of this problem and thank you for looking into this with me
tensorman_issues.txt
.

[mmstick]

Did you also install the system dependency for that? Not just the Python library.

[veirs]

I suspect you are pointing me toward something that the soundfile python library must use to access files. I don't know what the soundfile python library might 'depend' upon. I do know that it runs fine outside of the Tensorman container.

[veirs]

I think I have found it. I likely need something called PortAudio installed in the container. I will figure this out. Thanks for your help. If I get a simple solution, I will post it back here so that someone else might be able to use soundfile to read audio and then process data via Tensorman.