Introducing OhanASR: Open Armenian Speech Recognition Models
Blog post from Hugging Face
Metric-AI has introduced OhanASR, a family of open-weight Armenian speech recognition models named after a character in the Armenian epic Daredevils of Sassoun. The first release includes OhanASR-v0.1, a 2-billion-parameter multilingual model fine-tuned on 820 hours of Armenian speech for recorded audio, and OhanASR-Streaming-v0.1, adapted for incremental, real-time transcription. On the ArmBench-ASR benchmark, the standard model recorded a combined word error rate (WER) of 16.12% and character error rate (CER) of 3.49%, ranking second in WER and first in CER; the streaming model scored 19.07% WER and 4.47% CER, ranking tenth and ninth, respectively. Both models are available under the CC BY-NC 4.0 license for research, education, and other non-commercial uses, though they can struggle with formal and colloquial variants, mixed-language speech, and unclear or noisy audio. The team plans to improve accuracy, streaming latency, and performance across varied conditions, and reports that training was conducted on the LUMI supercomputer with access awarded by the EuroHPC Joint Undertaking.
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.