Add kokoro support

[daavoo]

Test in https://colab.research.google.com/github/mozilla-ai/document-to-podcast/blob/text-to-speech-model/demo/notebook.ipynb

[stefanfrench]

@daavoo When testing locally, I get ModuleNotFoundError: No module named 'phonemizer'

Do we need to update the dependencies?

[Kostis-S-Z]

Awesome stuff! I dont have many comments on the code.

As Stefan mentioned, there is the issue of updating the dependencies. Especially how to handle the espeak-ng dep.

And then there is another bigger question: How "fully" do we want to integrate kokoro? Or a better phrase is are we fully moving away from Oute and into kokoro? I am asking this due to the following issues:

1. What should we do regarding dependencies? Should we make outetts optional? Should we completely remove outetts support, as we did with parler and bark?

2. What should we do with testing? Should we add kokoro to test_model_loaders and test_text_to_speech?

3. Should we update all the docs references, specifically in step-by-step to reference kokoro instead of oute?

4. I dont think it makes sense for me to meticulously review the inference/kokoro/ files, as I am hoping we will soon replace them for the pip package 🤞

[Kostis-S-Z]

Also kokoro seems to pad each audio clip with at least 1 second silence and then the final complete podcast has a bit too much silence between speakers. I would update here like this:

        audio_np = stack_audio_segments(
            st.session_state.audio, speech_model.sample_rate, silence_pad=0.0
        )

[daavoo]

What I am thinking is:

• We integrate kokoro only for demo purposes (until a stable version is released in pypi)
  We handle dependencies either in the notebook or the Dockerfile used for the HF Spaces.
  We don't update README, docs or tests until pypi release.
  We do update references in the demo notebook and app.
• We keep outetts the default for the rest of the cases.

WDYT @stefanfrench @Kostis-S-Z

[stefanfrench]

What I am thinking is:

• We integrate kokoro only for demo purposes (until a stable version is released in pypi)
  We handle dependencies either in the notebook or the Dockerfile used for the HF Spaces.
  We don't update README, docs or tests until pypi release.
  We do update references in the demo notebook and app.
• We keep outetts the default for the rest of the cases.

WDYT @stefanfrench @Kostis-S-Z

Okay I'm comfortable with that

[Kostis-S-Z]

What I am thinking is:

* We integrate `kokoro` only for demo purposes (until a stable version is released in pypi)
  We handle dependencies either in the notebook or the Dockerfile used for the HF Spaces.
  We don't update README, docs or tests until pypi release.
  We do update references in the demo notebook and app.

* We keep `outetts` the default for the rest of the cases.

WDYT @stefanfrench @Kostis-S-Z

Works for me! So you need to revert the changes in demo/app.py, example_data/config.yaml, src/document_to_podcast/cli.py, src/document_to_podcast/config.py and maybe also push this and then merge?

[daavoo]

Works for me! So you need to revert the changes in demo/app.py, example_data/config.yaml, src/document_to_podcast/cli.py, src/document_to_podcast/config.py and maybe also push this and then merge?

demo/app.py is used in the HF space, needs to use kokoro there. Rest is done

[stefanfrench]

Thanks for making the changes.

Tested on Colab, local app, and CLI - works as expected. Approved.