tebinraouf/central-kurdish-tts

🗣️ Central Kurdish (Sorani) Text-to-Speech

A demo of `aranemini/central-kurdish-tts` — three single-speaker F5-TTS voices for Central Kurdish, developed in the TTS4All project at the JSALT 2025 workshop and at LIUM, Le Mans University.

How it works

The checkpoints are trained on phonemised text, not raw Arabic script. This Space reproduces the model repo's own infer.py frontend before synthesis:

1. 1.numeric range and long-number handling,
2. 2.asosoft.Normalize → asosoft.NormalizePunctuations → asosoft.Number2Word,
3. 3.asosoft.KurdishG2P, with ẍ → χ, ƹ → ¿, ḧ → ḥ.

The resulting phoneme string is shown next to the audio so you can see exactly what the model received. Each voice is conditioned on the prompt recording and transcription released alongside its checkpoint.

Runs on ZeroGPU; all three checkpoints are loaded at startup, so switching voices is free.

API / MCP

The Space exposes a /synthesize endpoint and doubles as an MCP server.

from gradio_client import Client

client = Client("tebinraouf/central-kurdish-tts")
audio, phonemes, seed = client.predict(
    text="سڵاو، بەخێربێیت بۆ کوردستان.",
    voice="studio-M — male, studio",
    speed=1.0,
    nfe_step=32,
    cfg_strength=2.0,
    cross_fade_duration=0.15,
    seed=-1,
    api_name="/synthesize",
)

License and citation

The model weights are released under CC BY-NC-ND 4.0: research and educational use only, no commercial use, no redistribution of modified versions. This Space loads the released checkpoints unmodified from the Hub. Please cite:

@inproceedings{mohammadamini-etal-2026-central,
  title = {Central Kurdish Text-to-Speech and Its Application in Speech-to-Text Translation},
  author = {Mohammadamini, Mohammad and Shamsi, Meysam and Tahon, Marie},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference (LREC 2026)},
  year = {2026},
  pages = {664--673},
  publisher = {European Language Resources Association (ELRA)},
  doi = {10.63317/4hfwowidu34u}
}