Zirek

Big vendors don't localize for under-served languages. The commercial market for any single one rarely justifies the investment, leaving speakers of hundreds of languages without screen reader support or local text-to-speech—regardless of community size. Zirek exists to close that gap through open-source infrastructure rather than commercial business cases.

Kurmanji Kurdish—spoken by 15–20 million people with no production-quality screen reader support today—is where this starts, not where it ends. The architecture is deliberately built to generalize:

A first-generation Kurmanji voice model already exists, trained with two Zarok TV voice actors, proving the pipeline works end-to-end: from raw studio audio to a functional NVDA voice.

Kurmanji's NVDA interface translation has remained open and incomplete since 2018—not due to tooling limitations, but because one-off volunteer enthusiasm rarely translates into sustained maintenance. dengjen-werger targets that exact bottleneck, starting with Kurmanji before expanding outward.

Built today by a solo maintainer on evenings and weekends alongside a full-time engineering role, Zirek is entirely self-funded. While an NLnet Foundation grant application is in progress to fund specific Kurmanji development milestones, this Collective supports the broader infrastructure that grants cannot cover: