Vekol

Vekol-TTS (Sorani, edge)

Central Kurdish (Sorani) text-to-speech that runs offline on CPU. A small ONNX model (~77 MB) that turns Sorani text into a 22.05 kHz WAV, faster than real time on a laptop CPU. Part of the Vekol hub by Revge.

  • Model: vekol-tts-ckb-edge
  • Language: Central Kurdish / Sorani (ckb), Arabic script
  • Voice: one male Sorani voice
  • Architecture: VITS (Piper recipe), fine-tuned two-stage, exported to ONNX
  • Size: model.onnx ≈ 77 MB, 22.05 kHz
  • Runtime: ONNX Runtime (CPU)

License

CC-BY-NC 4.0 (non-commercial). Built with the Piper VITS recipe and trained with audio from the Vekol server model (built on Coqui XTTS, non-commercial), so it inherits non-commercial terms. See NOTICE. Commercial use needs a model retrained from a permissively-licensed base on permissively-licensed audio.

Usage

The simplest path is the vekol_tts.py helper from the GitHub repo, which downloads this model and handles text normalization (variant folding, number reading, no dropped letters):

pip install onnxruntime numpy scipy huggingface_hub
python3 vekol_tts.py "ئەمڕۆ کەشەکە خۆشە" out.wav

It's a standard Piper/VITS ONNX graph, so it also runs directly via ONNX Runtime using the model.onnx.json phoneme map (inputs: input, input_lengths, scales, sid), or through the Piper / sherpa-onnx runtimes.

Notes

  • The character map uses Sorani keheh ک (U+06A9) natively — no remapping needed. The helper folds typed variants (كک, ھه, ةە, ىی, hamza, ZWNJ/tatweel) so nothing is dropped, and reads numbers as Kurdish words.
  • Output is 22.05 kHz: clear and accurate, intended for on-device/CPU use.
  • Runs through ONNX Runtime on many platforms (C++, Rust, Go, Android, iOS, Raspberry Pi, e.g. via sherpa-onnx or Piper).

Links

Citation

@software{vekol_tts_ckb_edge,
  title        = {Vekol-TTS: Sorani (Central Kurdish) on-device TTS},
  author       = {Shvan, Darvan},
  organization = {Revge},
  year         = {2026},
  url          = {https://github.com/Revge/vekol-tts-ckb-edge}
}

Built by Darvan Shvan at Revge, part of the Vekol hub.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support