Human-sounding local/offline Urdu TTS for Home Assistant? (Piper has no Urdu voice + XTTS-v2 doesn’t support Urdu officially)

Configuration Voice Assistant
1

Hi everyone,
I’m building a fully local/offline voice setup for Home Assistant and I’m stuck on Urdu TTS.

What I need

  • :white_check_mark: Local/offline only (no cloud APIs)
  • :white_check_mark: Urdu (اردو) voice
  • :white_check_mark: As human-like / natural as possible (not robotic)

What I already found

  • Piper: I can’t find any official Urdu voice/model for it.
  • Coqui XTTS-v2: official language list doesn’t include Urdu (it lists languages like English, Arabic, Hindi, etc., but not Urdu).

What I’m asking the community

  1. Is there any good offline Urdu neural TTS that you’ve actually used (VITS, Glow-TTS, etc.)?
  2. Any working community Urdu models that sound decent (HuggingFace models, GitHub projects, etc.) — please share exact names/links.
  3. Best way to plug it into HA? I’m okay with Docker / Python / Wyoming protocol if needed (Wyoming integration exists in HA).

I know eSpeak-ng supports Urdu, but it tends to sound robotic — I’m looking for something more natural.

2

So you can prepare a dataset and create a voice for Piper.
This seems like the most optimal solution.

If you find an existing project for this language, you will have to wrap it in the Wyoming protocol, after which the server can be launched in any convenient way - directly on the host, venv/uv, docker