ಕೊಡವ ತಕ್ಕ್ · ನುಡಿಯೇ ಬರಹ

The recording is the text.

Kodava Takk has no script of its own — it lives in the voice. We are building a community-owned archive of that voice, and teaching AI to listen, speak, and pass it on. UNESCO calls our language definitely endangered. We call it ours to keep.

● Give your voice — 30 seconds countsRead the manifesto
The Speaking Flame — the thook bolcha lamp radiating waveform arcs over the hills of Kodagu
0
Voices contributed
0h 00m
Thakk recorded
0
Recordings in the archive
The corpus is open — be among the first voices.Every recording is an heirloom: one elder, one story, one blessing at a time.The corpus is open — be among the first voices.Every recording is an heirloom: one elder, one story, one blessing at a time.

Now open · Level 0

The first hundred hours


Every language AI begins the same way: with recordings. Level 0 is the founding drive of the Kodava Thakk corpus — voices from every village, every okka, both dialects, home and diaspora. The dial above moves with every clip received.

Record a story, a song, a proverb, a blessing, or just answer a prompt about your morning coffee. Thirty seconds from a phone is a real contribution; an hour with an elder is an heirloom.

How it works

  1. Record in the browser, or upload a file — no app needed.
  2. Tell us who is speaking: village, okka, dialect, age.
  3. Choose your consent — archive, research, model training, public listening.
  4. Your clip lands in the community tracker, described and safe.
  5. AI pre-sorts and pre-transcribes; people verify. The corpus grows.

The plan

Four layers, one purpose


Layer 0 · Guardianship

The community owns the data

A Kodava data guardianship licence, modelled on the Māori Kaitiakitanga licence: contributors keep moral ownership, a community trust is custodian, and the corpus can never be sold. A council of elders, teachers, and technologists stewards every decision.

Layer 1 · The corpus

1,000 hours of living Thakk

Elder and Palame sessions recorded in depth; crowd voice drives for breadth; radio and family archives digitized; the Pattole Palame and dictionaries brought online. Both dialects, Mendele and Kiggat, by design.

Layer 2 · The models

AI that hears and speaks Thakk

Speech recognition first, fine-tuned from open Dravidian models. Then text-to-speech from consenting voice donors, then translation between Thakk, Kannada, and English. The app learns from your voice.

Layer 3 · The products

Someone patient to talk to

A public listening archive, a talking dictionary, classroom packs — and Ainmane, the speech-to-speech companion that lets a learner speak imperfect Thakk and be answered, gently, in Thakk.

See the full 36-month roadmap and milestones →

Join in

This is done by us, together

Kodava Thakk runs on the Kodagu.ai community — the same open network behind Aane Alert, Nela, and Manabala. Recordists, transcribers, singers, linguists, developers, and every fluent grandparent: there is a seat for you.

Ways to contributeJoin the Kodagu.ai community