Audio nodes (text2speech, speech2text)

[ZanSara]

Parallel Haystack PRs - deepset-ai/haystack#3229 deepset-ai/haystack#2676

Due to the (rather fragile) additional deps introduced into Haystack with the addition of Speech nodes, this PR creates the equivalent external nodes, to simplify maintenance of both projects.

Note: after this change, extra nodes will be to be imported by the script loading the YAML pipeline, in order to be found. So for example:

from haystack import Pipeline
from text2speech import AnswerToSpeech

pipe = Pipeline.load_from_yaml("pipeline.haystack-pipeline.yml")

Components ported:

• Nodes:
  • text2speech: AnswerToSpeech, DocumentToSpeech and relative base classes
  • speech2text: Wav2VecTranscriber, AeneasTranscriptAligner and relative base classes
• Utility classes
• Relevant primitives: SpeedDocument and SpeechAnswer (removed)
• Relative errors: Text2SpeechNodeError, Speech2TextNodeError
• print_answers --> print_audio_answers (removed)
• Test for nodes, utilities, primitives, and their samples
• Tutorial
• Dependency blocks from pyproject.toml
• JSON Schemas
• CI - now each package has their own, namespaced in .github/workflows
• Pre-commit hooks - now each package has their own
  • These new pre-commit also run mypy and the YAML generation quite quickly!
  • PyLint couldn't be made working
• mypy, pylint, pytest, black... configurations in pyproject.toml

Open questions:

• IDE YAML Validation: seems to be working but I want to make more in-depth tests
• Release process: synced with haystack?