Learning
ShabdPatra
Flip-card explorer for alphabets, numerals, words, phrases, idioms and proverbs across 14 languages of Nepal.
Open toolTools
The toolchain that turns recordings into working models.
Every dataset we deliver passes through the same set of tools — built for low-resource languages, unreliable connectivity, and community reviewers who are not engineers.
Collection
Voice Capture Studio
Browser recording workflow with prompt scripts, live level monitoring, retakes and per-clip consent capture.
Available to partnersAnnotation
Transcription & Alignment Desk
Audio-to-text workbench with timestamp alignment, dual-review validation and dialect tagging.
Available to partnersData
Corpus Manager
Versioned datasets with speaker metadata, license terms, quality scores and export to training formats.
Available to partnersEvaluation
Benchmark Runner
Standard ASR (WER/CER) and translation (BLEU/chrF) evaluation on held-out sets for every supported language.
Available to partnersDeployment
Chatbot Sandbox
Test written and spoken conversation with fine-tuned models before shipping them into a product.
Available to partnersPipeline view
Where each tool sits
Collect
- Voice Capture Studio
- ShabdPatra
Refine
- Transcription & Alignment Desk
- Corpus Manager
Ship
- Benchmark Runner
- Chatbot Sandbox
Want access to the toolchain?
Partners and research groups can run their own collection programmes on Sabdakunja tooling, with our team supporting quality review.
Request access