is Forced Alignment available on prebuilt Docker images?

[mgs96]

Hello!

I'm working with the Forced aligner and so far everything has been positive!

Now I wanted to use the prebuilt docker images provided by Nvidia, but when exploring the images I don't see any reference to the Forced Aligner script I used on the Google Colab example you provide.

Is this expected? Do I have to use the pytorch:23.10-py3 Docker image and install NeMo from GitHub?

Thanks in advance.

[erastorgueva-nv]

Hi! If you want use NeMo, you can either use the PyTorch container and install NeMo on it, or use a NeMo speech container. There are instructions for both in the README, e.g. here. AFAIK, neither type of container contains any of the tools from this directory. However, all of those tools are essentially relatively lightweight Python scripts that use NeMo (and maybe some other dependencies that are mentioned in their requirements files).

So if I were you, I would set up one of the Docker containers and make sure NeMo is installed, then git clone the NeMo repository so that you have access to the NFA scripts here: https://github.com/NVIDIA/NeMo/tree/main/tools/nemo_forced_aligner.

Then you should be able to use NFA inside the container by calling the <path to your NeMo location>/tools/nemo_forced_aligner/align.py script.

[github-actions[bot]]

This issue is stale because it has been open for 30 days with no activity. Remove stale label or comment or this will be closed in 7 days.

[erastorgueva-nv]

Hi, marking as closed since I believe the issue is resolved.