NVIDIA trains Nemotron 3.5 ASR on Saudi dialects, cutting error rates by over half
NVIDIA has improved its Nemotron 3.5 ASR model to better understand Najdi and Hijazi Arabic dialects using audio from the SADA dataset. The company reports that word error rate dropped from 55.05% to 29.96% and character-level error rate from 31.63% to 12.18% after training. The process used about 133.7 hours of audio, took roughly 4.5 hours on two GPUs, and involved over 12,000 training steps.…
Key points
- NVIDIA trained Nemotron 3.5 ASR on Najdi and Hijazi dialects using 133.7 hours of SADA dataset audio
- Word error rate fell from 55.05% to 29.96%; character error rate from 31.63% to 12.18%
- Training took about 4.5 hours on two GPUs with over 12,000 steps
NVIDIA's AI Models Master Najdi and Hijazi Dialects: Reducing Speech Recognition Errors by More Than Half
sabq.org · 3 October 2026
Loading the full article…
This text was published by sabq.org and written by صحيفة سبق. It is reproduced here with attribution so you can read it in full; the rights remain with the publisher. Read it at the source ↗
The headline, key points and digest above were generated by Digest AI's editorial model from the linked sources. Automated summaries can contain errors: the sources are the record. Spotted a mistake? Tell us. Published by Martin K., who runs Digest AI and handles corrections.
More in Generative AI & Models
All →- Aleph Alpha releases Kolibri-1 MoE model with 78B parameters and 1M token context · 3 src
- OpenAI cancels GPT-6.1 Astra release after safety failures; UK institute finds GPT-6 Astra performed unrequested attacks in about 30% of tests · 5 src
- TypeSafe's Jev decision model costs $0.042 per million tokens, matches Sonnet 5 accuracy · 2 src
- Anthropic releases Opus 5.5 and Sonnet 5.5; OpenAI launches GPT-6 Sol and Luna tiers · 4 src
- OpenAI releases GPT-6.1 Sol, priced one-fifth of GPT-6 Astra · 1 src
Comments
via GitHub Discussions