Instructions to use zakumi/awavote-models with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers.js
How to use zakumi/awavote-models with Transformers.js:
// npm i @huggingface/transformers import { pipeline } from '@huggingface/transformers'; // Allocate pipeline const pipe = await pipeline('automatic-speech-recognition', 'zakumi/awavote-models');
AwaVote on-device speech models β Powered by Awarri
int8 ONNX conversions of the N-ATLaS speech-recognition models, used on the phone (in the browser, offline) by AwaVote, an election-transparency app built for the KTech Fest 3.0 hackathon. Source code: https://github.com/ktechfest2/TEAM-11-3.0
| Folder | Model | Source |
|---|---|---|
natlas-asr-yo-int8 |
AwaVote Yoruba ASR (int8) Powered by Awarri | NCAIR1/Yoruba-ASR |
natlas-asr-ha-int8 |
AwaVote Hausa ASR (int8) Powered by Awarri | NCAIR1/Hausa-ASR |
natlas-asr-ig-int8 |
AwaVote Igbo ASR (int8) Powered by Awarri | NCAIR1/Igbo-ASR |
natlas-asr-en-int8 |
AwaVote Nigerian English ASR (int8) Powered by Awarri | NCAIR1/NigerianAccentedEnglish |
The N-ATLaS models are developed by Awarri Technologies in partnership with the Federal
Government of Nigeria, and are licensed under the N-ATLaS Open-Source Research and Innovation
License (see each source model page). Each folder carries its NOTICE.txt. Use is subject to that
licence, including its limits on active users.
Conversion: Whisper-small fine-tunes exported to ONNX and quantised to int8, with a shortened
encoder window for short voice commands (ai/speech_to_text/ in the AwaVote repository).
registry.json lists every file with its size and SHA-256.