open Nepali AI — built in Kathmandu

about| work| mission| huggingface| contact

ampixa labs

We are ampixa labs — a Nepal-based research lab building open speech, language, and evaluation infrastructure for the Nepali language.

Most of the world's major AI systems do not understand Nepali well, and almost none speak it back fluently. The data, benchmarks, and tooling needed to fix that simply do not exist in the open. We build that.

Open code. Open data. Open evaluation. Built in Kathmandu, for everyone working on Nepali AI.

what we build

NepTTS-Bench — an open benchmark for Nepali text-to-speech, with reference recordings, evaluation pipelines, and human MOS ratings across leading systems. Dataset on huggingface.

Voice corpus — a growing crowd-sourced library of Nepali speech across dialects and speakers, captured under permissive licenses for research and product. Contribute at tts.ampixa.com/speak.

Human evaluation — a rating platform where native Nepali speakers compare TTS systems via MOS, A/B preferences, and minimal-pair tests. The ground truth for everything we publish. See tts.ampixa.com/rating.

mission

Democratize Nepali AI.

So the next billion people speak to AI in their own voice — not a translation, not a workaround, but the real thing.

Every benchmark we publish, every dataset we collect, and every model we release is meant to lower the floor for everyone working on Nepali AI — PhD students, startup founders, government services, hobbyists. Anyone.

find us

about the company

AMPIXA LABS Pvt. Ltd. is a privately held research and engineering company incorporated in Nepal under the Office of the Company Registrar. We operate from Kathmandu and publish our work under permissive open-source and open-data licenses.

© 2026 AMPIXA LABS Pvt. Ltd. — Kathmandu, Nepal.
Devanagari wordmark uses "8-bit devanagari" by colonelhathii (CC BY-NC 3.0).