Add support for audio queries

[gkennickell]

Adds audio support queries based on #233 updated to work with latest. This is a simplified/refactored version, discussion in: #577.

The atari_env.py support in the initial commit is not final, stubbed in for feedback purposes.

[pseudo-rnd-thoughts]

I think looks good. It would be good to do some testing with an actual algorithm.

For the python stage, it wouldn't be a "separate" return value, rather I think we should do a Dictionary of {"image": image_obs, "sound": sound_obs} if sound_obs==True.

We will need to update the observation_space if we are doing that. I'm happy to make some of those changes if you are uncertain.

@jjshoots Do you have any thoughts?

[gkennickell]

It would be good to do some testing with an actual algorithm.

I put together a simple algorithm which runs the image and sound data through separate feature encoders, concats, and then pushes through a simple conv network. Posting results for comparing sound_obs v no_sound_obs across a handful of atari games.

Additionally, posting a video generated from the per-frame image+sound data for breakout.

breakout.mp4

I'm happy to make some of those changes if you are uncertain.

I'm not as familiar with the gym api, so would definitely appreciate help integrating the feature in the cleanest way. I've amended the commit to remove the draft portion of the changes to env.py.

[pseudo-rnd-thoughts]

I'm not as familiar with the gym api, so would definitely appreciate help integrating the feature in the cleanest way. I've amended the commit to remove the draft portion of the changes to env.py.

No worries, I have added the relevant code.
Could you add the following code to the end of the tests/python/test_atari_env.py

def test_sound_obs():
    env = gymnasium.make("ALE/MsPacman-v5", sound_obs=True)

    with warnings.catch_warnings(record=True) as caught_warnings:
        check_env(env.unwrapped, skip_render_check=True)

    assert caught_warnings == [], [caught.message.args[0] for caught in caught_warnings]

Another note is our pre-commit raised the following issue. Could you add this.

check that executables have shebangs.....................................Failed
- hook id: check-executables-have-shebangs
- exit code: 1

src/ale/common/SoundRaw.cxx: marked executable but has no (or invalid) shebang!
  If it isn't supposed to be executable, try: `chmod -x src/ale/common/SoundRaw.cxx`
  If on Windows, you may also need to: `git add --chmod=-x src/ale/common/SoundRaw.cxx`
  If it is supposed to be executable, double-check its shebang.
src/ale/common/SoundRaw.hxx: marked executable but has no (or invalid) shebang!
  If it isn't supposed to be executable, try: `chmod -x src/ale/common/SoundRaw.hxx`
  If on Windows, you may also need to: `git add --chmod=-x src/ale/common/SoundRaw.hxx`
  If it is supposed to be executable, double-check its shebang.

Otherwise, I suspect that we might be close to ready to merge.

[pseudo-rnd-thoughts]

@gkennickell Sorry, one last change, could you add some documentation to docs/cpp_interface.md on accessing the audio queries.

Then we should be good to merge

Also could you run pip install pre-commit and pre-commit run --all-files in the project root

[gkennickell]

Added simple documentation in docs/cpp_interface.md + fixed pre-commit tests. Please let me know if more documentation is warranted (for instance, in environments.md) or minimal is a good start.
And, thanks for the help on pushing this through.

[pseudo-rnd-thoughts]

Looks good, I'll add more documentation later