README.md

metadata

license: cc-by-4.0
language:
  - am
  - om
  - ti
tags:
  - speech
  - text-to-speech
  - amharic
  - ethiopia
  - samples
  - azemari
pretty_name: Azemari — generated samples

Azemari — generated samples

Audio samples generated by Azemari — our Ethiopian multilingual voice (TTS): Amharic, Afaan Oromoo and Tigrinya with emotion control and voice cloning.

Groups:

• Root files: the final model — emotion set (happy, sad, angry, afraid, calm, disgusted, melancholic, surprised), voice clones (clone_*), dialect control (shewa_*), and multi-language tests (am_* Amharic, om_* Afaan Oromoo, ti_* Tigrinya)
• train-v2-interim/best/ — interim v2 checkpoint samples
• v3-step57500/ — v3 checkpoint samples (emotions + neutral)

The model itself: Lab-et/azemari — complete, self-contained checkpoint, ready to run.

Generated by lab.et — the AI lab of Ethiopia. Our other releases: amharic-corpus (408.5h open Amharic speech) and omnivoice-amharic.