Voice AI is still largely built English-first. We built ours for Amharic and Afaan Oromo.
Twenty-eight voices, not one demo voice
Addis AI Voices 2 launched with 28 voices for Amharic and Afaan Oromo. They are available through the Addis AI platform and the production API.
A catalog matters because products do not all need the same delivery. A support line, a research reader, a commercial, and a narrated lesson should not be forced through one voice profile.
Voice selection is part of the product
The API exposes the current voice catalog with language, gender, style, tags, a preview, and availability. Developers can search the catalog, let a user preview options, and store the selected voice ID for generation.
Availability comes from the live catalog. That keeps an integration tied to voices that can actually be generated instead of a static list copied from documentation.
Discover
Query voices by language and style.
Preview
Listen before selecting a voice.
Estimate
Check duration and cost before generation.
Generate
Create and save a durable audio clip.
Built for a production request
Voices 2 returns clip metadata and a signed audio URL. It supports MP3, WAV, and PCM output formats, along with request IDs for retries and clip management through the SDK.
Those details are not launch decoration. They are the parts a developer needs when speech becomes a real output in a customer workflow.
Try the voices
The live platform has the voice catalog and previews. The documentation covers discovery, estimates, generation, and migration from the earlier voice API.