1. Train
Listen and speak practice with meaning in Nepali and English. Phrasebooks stay usable offline. Real language tracks unlock with community partners — not empty marketing lists.
Open backbone · Nepal · Mother tongues
Practice oral languages.
Record with guardians.
Keep communities in control.
A long-term, open-source program to help Nepal’s spoken languages stay alive — starting with a mobile training app where young people learn and document speech, including with parents and grandparents.
Learn ↔ Contribute ↔ Transmit Private by default Partner-gated languages
Why this exists
Nepal has well over a hundred mother tongues. Many are spoken far more than written. Young people move toward Nepali and English; elders hold the deepest fluency. Documentation projects and NGOs often rebuild tools from scratch every grant cycle.
We are building shared rails: offline-friendly mobile practice, consented self- and guardian recording, dialect and origin notes, and open publication only when communities choose it.
Not extractive AI
Open software does not mean every voice recording is free to scrape. Clips default to private. Open and research tiers require explicit choice and steward oversight. Family stories are not a public feed.
How it works
2. Record with guardians
Young people operate the phone; elders speak. Dual consent is required. Speakers choose private, research, or open use. Refusal is always allowed.
3. Publish carefully
Only steward-approved open material goes to public Hugging Face packs. Research speech can be gated. Private vault material stays off the public internet.
Mobile app (early)
Nepal Oral Train is an Expo app in active development. Today it ships a demo language track so anyone can try lessons, consent flows, self-record, and guardian sessions without a real community partner.
- Demo lesson + phrasebook + language origin screen
- Operator consent (saved on device)
- Self-record with license choice
- Guardian dual-consent session
- Local queue (no automatic cloud upload yet)
Source:
apps/train
on GitHub. Run from that folder with Expo Go after
npm install and npm start.
Phone crashes during early previews are being hardened (navigation safety, permissions). Prefer a development build or latest main branch over stale QR links.
Open data on Hugging Face
Public and research artifacts publish under AkAiNp — not from every phone upload.
Demo
Synthetic fixtures for tools and UI. Never real speakers.
nepal-oral-demo →Open packs
Educational packs and seed audio after steward publish (starts empty).
nepal-oral-packs-open →Research (gated)
Research-licensed speech only. Access requests and partner terms.
nepal-oral-speech-research →For NGOs & language communities
You bring trust, seed phrases, and a named steward. We bring open mobile tooling, consent patterns, and pack standards so projects outlive a single grant.
- Hosted track, self-host path (roadmap), or content-only partnership
- Languages unlock with MoU-lite + seed content — not hard-coded hype
- Youth–elder recording protocol designed for real homes
- Metrics that matter: sessions, consent, pack quality — not vanity AI scores
Start a conversation via GitHub Issues on the repository, or contact AKT Tech through the org listed on GitHub.
Contribute
Code & design
Help with the Expo app, packaging, accessibility, and offline UX.
Read CONTRIBUTING.md on GitHub. Do not open PRs with
real community audio.
Community & languages
Introduce stewards, review consent language translations, or partner on a pilot track. Language content follows steward process — not random dumps into git.
A decade-scale public good
Downstream speech models and multilingual tools only come after consented corpora and community trust. Year one is the backbone: practice, intergenerational capture, and open rails NGOs can share.
Star & follow on GitHub