About this tool

What it is

A native-speaker validation form for the project's Nepali grapheme-to-phoneme (G2P) gold lexicon. The lexicon underpins every TTS evaluation we publish.

The frontend produces candidate pronunciations from rules and from a source lexicon (Google's language-resources/ne, CC BY 4.0). Native speakers and linguists validate those candidates one word at a time. Validated entries become the project's gold lexicon.

Why we use IPA-style labels and not raw IPA

Typing IPA in a browser is hard. The project's labels (ax, tx, dz, etc.) are ASCII, easy to type, and unambiguously map to a single IPA equivalent. The cheatsheet has the full table.

Why we collapse some orthographic distinctions in default mode

The default mode is spoken_nepali. Per the primary references (Khatiwada 2009, Regmi 2025):

श, ष, स all read as s in spoken Nepali.
ण and ञ both read as n.
ी (long ī) reads as i; ू (long ū) reads as u — spoken Nepali has no length contrast.

If you think a specific word needs the careful Sanskritized pronunciation, please note that in the comments field; don't try to encode it in default-mode phones.

What the form does to your decisions

Saves them in your browser's local storage as you go (so you can stop and resume).
When you click Export decisions as TSV, you get a single file with all your decisions plus your name.
You email or upload that file to the project. We run a script that validates phones against the v1.0 inventory and appends accepted entries to gold_lexicon.tsv.

Any phone string that doesn't validate (e.g., a typo'd label) is flagged for follow-up before promotion. Nothing reaches the gold without passing the inventory check.

Privacy and licensing

Your reviewer name is recorded with each accepted entry as reviewer = <name>.
Decisions become part of the project's open data. We release the gold lexicon under a permissive open licence on HuggingFace.
If you want your contributions credited differently or anonymously, let us know in advance.

Source-code references

The sources behind the v1.0 phone inventory and rules:

Khatiwada, R. (2009). “Nepali.” Journal of the IPA, 39(3): 373–380. The IPA Handbook illustration of Nepali.
Khatiwada, R. (2019). “Retroflexion in Nepali.” Gipan 4: 19–29. Direct palatographic study, 9 speakers.
Regmi, B. N. (2025). “Variation in Nepali Verb Roots: A Phonological Perspective.” Gipan 7(1): 121–135.
Pokharel, M. P. (1989). Experimental Analysis of Nepali Sound System. PhD thesis, University of Pune.

Contact

Questions about a specific entry, the workflow, or wanting to contribute: hello@ampixa.com.