571 words · 19 sentences · 0 voice clips — approved so far
A Tangkhul Naga AI,
built by its speakers.
Words, sentences, and voice — collected by the people who speak it, reviewed by humans, versioned like code, and exported ready for Whisper, LLM, and TTS training.
The ledger
live · open for contribution| Phase | Approved | Pool | Progress |
|---|---|---|---|
| Words | 571 | 44,254 | 1% |
| Sentences | 19 | 38,551 | 0% |
| Voice clips | 0 | 0 | 0% |
43,683 word prompts and 38,532 sentence prompts are still open. Training starts when you decide the dataset is full. Sign up →
Three phases, one dataset
Start with words, build up to natural speech. Each phase feeds the next.
-
01
Words
English prompts → Tangkhul with meaning, part of speech, IPA, and real usage examples. Variations encouraged.
-
02
Sentences
Sentences and mini-dialogues with register tags, so the model learns how Tangkhul is actually spoken.
-
03
Voice
Record accepted sentences. Every clip is auto-checked for clarity (SNR) and Whisper round-trip accuracy (WER).