Nepali pronunciation review

Native speakers and linguists validate proposed Nepali pronunciations one word at a time. Decisions are append-only; promoted entries become the project's gold lexicon, used to train and evaluate Nepali TTS.

No login. No internet needed once a batch is loaded. Decisions auto-save in your browser; export TSV when done.

open batches

loading…

how it works

Pick a batch above. Each batch is ~50–500 Nepali words.
For each word, see the suggested phones, an IPA hint, and a few example sentences from a Nepali corpus.
Decide: Accept, Correct, Reject, or Skip. Keyboard shortcuts 1 / 2 / 3 / 4.
When done, click Export decisions as TSV. Email the file back to hello@ampixa.com or upload to the project's review channel.

Phone-label reference: cheatsheet. Full workflow doc: about this tool.

two reviewer roles

Role	Typical batch	What you catch
Native speaker	100–500 high-frequency unknowns	Pronunciations the rule-based G2P had to guess at — usually inflected forms, named entities, and recent loanwords.
Linguist	200–1000 mixed (audit-style)	Systemic phonological issues — wrong sibilant collapse, schwa retention rules, conjunct boundary handling.

what happens to your decisions

Accepted / corrected entries are appended to the project's gold_lexicon.tsv with your name as reviewer and a source = native_review tag. They are released under the project's open licence.
Rejected entries are routed to rejected_entries.tsv so we don't suggest them again.
Skipped entries stay in the candidate pool for future review sessions.

Phones in your accepted entries are validated against the project's v1.0 phone inventory before promotion. Typos in the corrections are caught at promotion time, not silently merged.