Open Nepali text-to-speech
new
— nine tiny neural voices, six languages, live in your browser
and on a $3 chip. try it →
Open Nepali text-to-speech
new
— nine tiny neural voices, six languages, live in your browser
and on a $3 chip. try it →
| /sanoTTS |
— nine tiny neural voices (745k–1.8M params) in
six languages: English, Nepali, Hindi, Vietnamese, Indonesian,
Chinese. Synthesizes live in your browser, runs faster than real time on
a $3 ESP32. code,
models.
pip install sanotts
new
|
| /kala |
Kala TTS
— the first open-source Nepali VITS TTS with its own
Devanagari G2P, no eSpeak. 5 voices, 50× real-time on CPU.
Open model,
code, and
training recipe.
pip install kala-tts
|
| /speak | Voice recorder — record reference audio for the open Nepali voice corpus. Anyone can contribute. Sentences come from NepTTS-Bench. |
| NepTTS-Bench | huggingface.co/datasets/ampixa/neptts-bench — 365 sentences, minimal pairs, prosody, robustness, plus reference audio and MOS ratings. |
| Voice corpus | Open-license Nepali speech recordings. Released in tagged versions on HuggingFace. |
| G2P inventory | Project-owned phone inventory and lexicon for Nepali, grounded in Khatiwada (2009), Pokharel (1989), Regmi (2025). Public after the first reviewed gold lexicon ships. |
© 2026 AMPIXA LABS Pvt. Ltd. — Kathmandu, Nepal.
Devanagari wordmark uses
"8-bit devanagari"
by colonelhathii (CC BY-NC 3.0).