[FORK] Diversion

[nicandris]

Hey, while playing around with this trying to make it fit my needs (thank you so much btw! 🙇 ) I diverged quite a lot from the origin. I didn't want to push all these changes back as one or many PRs so I had claude create this doc. https://github.com/nicandris/Kokoro-TTS/blob/main/DIVERSION.md

You can cherry pick what you want/need or you can let me know and i can create individual PRs for them. Thank you again for the project!!

[beecho01]

Hey @nicandris,
Thanks for this! I have had a review and I think you ran this before I did some changes 4 days ago. Must have missed it by a few hours! We you be open to re-running the diversion doc again and I will be happy to review this. There are definitely some bits in there that would make the integration and repo, more mature.
Big thanks!

[beecho01]

@nicandris I have just pushed the latest release 2026.08.19. Feel free to have a look at that and see if we have both converged on some quality of life improvements.

[nicandris]

Thanks for taking the time to look, and for cutting 2026.08.19.

Good news on the timing worry: 48014b9 (manifest 2026.08.15) turns out to be the
merge-base between our two repos, so your changes from that day were already in scope
when the doc was written. Every finding and line reference still applies to your current main,
and I've updated the doc header to say which commit it's pinned to.

For §1 I re-ran your own get_persona_display_name / get_technical_persona_name as they
stand at da02338, across all 54 entries in PERSONA_MAPPINGS with both filters set.
Five voices don't survive the round trip:

Picked	Stored	You get
hf_alpha (Hindi)	jf_alpha	Japanese
em_santa (Spanish)	am_santa	American English
pm_santa (Brazilian Portuguese)	am_santa	American English
pf_dora (Brazilian Portuguese)	ef_dora	Spanish
pm_alex (Brazilian Portuguese)	em_alex	Spanish

Brazilian Portuguese has three voices, and all three are in that list.

On converging — not yet, as far as I can see. These are the three I'd still flag, all
present at da02338:

• aiohttp.ClientSession built per request at four sites (tts.py:240, tts.py:327,
  config_flow.py:146, config_flow.py:205) rather than Home Assistant's
  async_get_clientsession.
• HTTP 401 only ever becomes a message string (tts.py:154, config_flow.py:208), so
  nothing raises ConfigEntryAuthFailed and the async_step_reauth flow you already
  wrote can't be reached.
• self._attr_unique_id = f"kokoro_tts_{name}" (tts.py:130), which collides when two
  entries point at different servers.

No rush on any of it, and no hard feelings if you'd rather write your own fixes — the doc
is there to be raided.