refactor: Remove AnswerToSpeech and DocumentToSpeech nodes

[silvanocerza]

Proposed Changes:

This PR completely removes AnswerToSpeech and DocumentToSpeech audio nodes and relative tests.

We decided to completely remove them directly as the nodes are not used as shown by our internal metrics.

This nodes will still be usable from the Haystack extras.

How did you test it?

I run tests locally.

Notes for the reviewer

Hold merging this until we publish the extra packages.

[julian-risch]

@agnieszka-m @silvanocerza What's the plan with the documentation? For example with this page: https://docs.haystack.deepset.ai/v1.15-unstable/docs/answer_to_speech
I'd say all the nodes in extras should get a link to the extras repo and a brief explanation that it's not part of Haystack core and how to install. Or will the docs be restructured with an extra section about extras?

[ZanSara]

@silvanocerza don't forget to remove Speech primitives:

https://github.com/deepset-ai/haystack/blob/main/haystack/schema.py#L283
https://github.com/deepset-ai/haystack/blob/main/haystack/schema.py#L481

and audio dependencies:

https://github.com/deepset-ai/haystack/blob/main/pyproject.toml#L149
https://github.com/deepset-ai/haystack/blob/main/pyproject.toml#L229
https://github.com/deepset-ai/haystack/blob/main/pyproject.toml#L233
https://github.com/deepset-ai/haystack/blob/main/pyproject.toml#L67

We also don't need libsndfile1 in CI anymore (like here https://github.com/deepset-ai/haystack/blob/main/.github/workflows/tests.yml#L461), however we do need ffmpeg for the Whisper node

[vblagoje]

Yes, good points @ZanSara I'll, in the meantime, make some minor changes to #4335 . Let's integrate this PR first and then #4335

[vblagoje]

Noice @silvanocerza

[agnieszka-m]

@julian-risch @silvanocerza so what's the plan for this "extras" repo? what's going to be there?

we definitely need some documentation around it.

But I assume the nodes moved to extras will still work as described in the current docs? just the installation is going to differ?
Maybe we should introduce Haystack core and Haystack extras in the haystack concepts section and then update the installation instructions.

Is this going to be in 1.15?

[silvanocerza]

But I assume the nodes moved to extras will still work as described in the current docs? just the installation is going to differ? Maybe we should introduce Haystack core and Haystack extras in the haystack concepts section and then update the installation instructions.

As of now they'll be only moved in the deepset-ai/haystack-extras repo and published as separate packages. Their logic will be left untouched but it's possible it might change in the future. 🤷

I like the idea of introducing the concepts of Haystack core and extras in the installation instructions.

Is this going to be in 1.15?

Yes, they'll be removed already in the next version.

[julian-risch]

Looks very good to me already, just a few tiny things that can also be removed:

• AudioNodeError in haystack/errors.py is unused now and should be removed.
• AnswerToSpeech and DocumentToSpeech safe_imports should be removed from haystack/nodes/init.py
• Searching for "audio" in the code base gives ContentTypes = Literal["text", "table", "image", "audio"] in haystack/schema.py and also some results in Embedder. We should add a link to the extras repo maybe, because it relies on it, for example here but looking just at Haystack, it's hard to understand why we still have it.

[julian-risch]

LGTM! 👍

[silvanocerza]

Rebased to fix conflicts.