Add universal phone recognition model - PhoneticXeus

[Shikhar-S]

What does this PR do?

This PR introduces PhoneticXeus, which is the state-of-the-art universal phone recognizer trained on 70+ languages and evaluated on ~100 languages. The model should have high utility for the linguistic, phonological and multilingual research community.
Since this is currently the best-performing multilingual phone recognition model available, I expect the integration here will also attract new users to HF.

The E-Branchformer encoder used here is architecturally distinct from existing models, and has been employed in top performing speech models (XEUS, OWSM v2+ etc). The implementation here (ported from espnet) could also serve as a foundation for future speech models in HF.

Code Agent Policy

I have read CONTRIBUTING.md.

• I confirm that this is not a pure code agent PR. I used coding agent for initial draft of this PR. Then I tested the output with original implementations, and manually reduced verbosity of the code generated.

Before submitting

• Did you read the contributor guideline,
  Pull Request section?
• Did you make sure to update the documentation with your changes? Here are the
  documentation guidelines, and
  here are tips on formatting docstrings.
• Did you write any new necessary tests?

[Rocketknight1]

Audio model I guess, albeit an unusual one, so cc @ebezzam @eustlb! Transcription to IPA is very niche (but very cool) though, so we may want to keep this as a remote code model unless we expect significant usage.

[github-actions]

[For maintainers] Suggested jobs to run (before merge)

run-slow: auto, phoneticxeus

[Shikhar-S]

Thanks @Rocketknight1 for the initial review!
I have added motivation for this PR to the description above. Additionally, integrating the model into HF ecosystem will lower the barrier for people to easily reproduce results and build on top.

[eustlb]

Hey @Shikhar-S, thanks a lot for your PR, really cool contribution!

I’d like to confirm what @Rocketknight1 mentioned above: this model is a better candidate for remote code, which will allow you and users to fully leverage the HF ecosystem. If it gains strong usage, it could become a great candidate for native support 🤗

Thanks a lot for your understanding! I’m happy to help with the remote code option if you need it

[Shikhar-S]

Hi @eustlb thanks for taking a look!
For adding it via remote code option, do I need to move the modeling code to my hf repo, and just keep the auto registrations here? Is there an example I can look at/or some documentation for this option? I will change the PR accordingly. Thanks!

edit: adding @Rocketknight1 to advise next action for me.