Read aloud interest check

[avifenesh]

I have a personal patch that reads aloud the response, sometimes when I'm too lazy to read it all. For normal voice it requires some ONNX model, like Kokoro, 200 MB, running on CPU. Otherwise a robotic unpleasent one that was built in the distro.
Any interest?
Can add it as opt-in, with a comfortable download model from HF for those who want it.
Will probably make it MCP for live conversation next.

[ilysenko]

Thanks, this sounds useful.

I think it should be an opt-in Linux Feature, not enabled by default. Users could enable it through linux-features/features.json.

Please avoid bundling a large ONNX model by default. An optional download/cache flow from Hugging Face sounds good, and the distro TTS fallback is fine too.

MCP/live conversation can come later; a simple read-aloud Linux Feature would be a good first step.

[avifenesh]

@ilysenko that's exactly the target, obviously not going to pack 200 MB to a user, and yeah, step by step, small chunks.

[avifenesh]

Opened #235 for the first opt-in Linux Feature slice: no bundled model, Hugging Face download/cache flow, local folder path, explicit read button, and no MCP yet.

[avifenesh]

I leave it open for the MCP, and later maybe conversation mode. Although the latter is a big speculation, cool challenge tho.

[avifenesh]

It is super cool actually. With the MCP, Codex can talk when it wants. I gave an instruction to update by voice. Codex runs a long session, i do my things. Once in a few minutes I get an update on what's going on in my earphones.
The repo is getting better than the original.

[avifenesh]

#242
Was very interesting, and now i really feel the term "vibe coding". Not even typing.
@ilysenko @Leay15 Please let me know if you try it, im genuinely interested how it feels.

[Leay15]

@avifenesh I'm trying to use the feature but I doesn't works, I see all speech icons but I don't hear anything.

I already activate it on settings but nothing occurs:

After clicking Download voice it changes the state but restarts afterwards.

🤔 Do I have follow some steps to making it work?

[avifenesh]

@Leay15 I noticed that there is also a setting in general. Can you try from there and let me know if it works? Maybe I made some location mistakes.

[Leay15]

@avifenesh It behaves the same. Both settings options share the same state, and changing them doesn't affect the behavior. I can see that everything is enabled, but the speech icon still doesn't work.

[avifenesh]

@Leay15 and did you try the manual? i want to know if its a matter of the download and the native, or an issue with the whole thing

[Leay15]

@avifenesh Yes, I ran bash linux-features/read-aloud/install-kokoro-runtime.sh, and it downloaded the models to the default location.

Am I missing something?