If you're reading this, you're probably interested in contributing to Vibravox. We're happy to have you here!
This repo is meant to be a collaborative project on several tasks related to speech processing recorded by body-conducted microphones. Here is the list of currently supported tasks:
Task Tag | Description |
---|---|
bwe |
Bandwidth Extension |
stp |
Speech to Phoneme |
spkv |
Speaker Verification |
If you're using the Vibravox dataset, you can add your model to one of the supported tasks. Here's how to do it:
- The entry point of your code should be the
run.py
script. - You must use the lightning datamodule corresponding to your task.
- You can create a new lightning module, some torch_modules or any other utils you need.
- For every created class, you should add the corresponding yaml file in the
configs
folder. - Finally, your method should be run with the following command:
python run.py lightning_datamodule=<task_tag> lightning_module=<your_model> <other_args>
Please open an issue to discuss it.
git clone
the repo and create a new branch withgit checkout -b <new_branch>
.- Make your changes, test them, commit them, and push them to your branch.
- You can open a pull request on GitHub when you're satisfied.
Note: Use the following naming convention for new branches : new_branch = <github_username>/<task_tag>/<feature|bug>/<branch_description>
Things don't need to be perfect for PRs to be opened, draft PRs are welcome.
All contributions to the source code should be documented.
Docstrings follow the Google format, have a look at other docstrings in the codebase for examples.