Sovereign speech technology

Language tools that stay with the community that owns the language.

KinSpeak builds speech recognition, natural voices, talking books and classroom tools for Indigenous language revitalisation. Everything runs on local hardware. No cloud services, no third parties, no data leaving community control.

What a community gets

One platform, delivered as working tools, owned outright by the community at handover.

📖

Talking books and dictionaries

Any dictionary, story or reader the community holds becomes a tap-to-hear talking edition, with the voice chosen by the community.

🎙️

A speech engine for the language

Recognition that writes the language down and synthesis that reads it aloud, trained for this language specifically, running on one local machine.

🏫

Classroom tools

Lesson builders, word games, pronunciation practice and progress tracking, built only from vocabulary verified against published grammars and speakers.

🤝

Workplace inductions

First-words modules for staff who work with the community, with recording, comparison and completion records for employers.

🗄️

Archive transcription

Legacy tapes and field recordings become searchable, time-aligned text, returned to the community as usable language material.

🎬

Media in language

Production tooling for bringing media into the language: timing, per-line quality checking and sound mixing around community voice actors, and synthetic narration for content the community owns.

How we work

Built first for a first language community in northern Australia, and designed so the whole platform can be stood up for any language.

Community ownership. Language data and models are community attributed and transfer to the community outright, on the community's decision.
Consent in code. A voice is only ever served when that speaker's documented consent is on record. The rule is enforced by the software, not by promise.
Honesty over polish. Every word served traces to a published grammar or a speaker recording. Words that cannot be verified are reported as unknown, never guessed.
Local by architecture. The whole system runs on hardware the community can physically hold. Nothing depends on a cloud subscription that can be switched off.

Who builds it

KinSpeak is built by a machine learning engineer with a long personal connection to northern Australia. The work is done under community protocols, with consent guided by people who hold that authority, and with the position that the technology is finished only when the community can run it on their own.

Working on a language? Demonstrations by invitation. Happy to talk to language centres, schools, communities and the organisations that employ people on their Country.

rod@kinspeak.org