Open speech and language models for Armenian: OhanASR, OhanTTS and Ayb
Three open model families, released together: Armenian speech recognition (batch and streaming), speech generation, and a large language model with vision.
Read
All posts
4
Introducing OhanASR v0.1
State-of-the-art open Armenian speech recognition — #1 by character error rate on ArmBench-ASR, with a streaming variant for real-time use.
Read

Introducing OhanTTS v0.1
Streaming Armenian text-to-speech with voice cloning — first audio in about a third of a second, in eight languages.
Read

Introducing Ayb v0.5
The first step towards sovereign Armenian AI: a state-of-the-art open LLM that rivals Gemini, GPT and Claude on Armenian, with early vision.
Read
Nothing here yet — check back soon.