# Nvidia reports faster, more accurate Saudi Arabic transcription in decoder test

_Published Thursday, October 1, 2026 at 2:06 AM EDT · AI · Latest · Tier 2 — Notable_

![Nvidia reports faster, more accurate Saudi Arabic transcription in decoder test — Primary](https://developer-blogs.nvidia.com/wp-content/uploads/2026/09/image3-18.webp)

Nvidia reports that changing the decoding configuration of a fine-tuned Nemotron speech recognition model improved Saudi Arabic transcription accuracy while reducing runtime in its test. Using NeMo's batched MALSD beam-search strategy, beam 4 lowered the word error rate from 29.96% to 28.81% at 0.59× the greedy runtime.

The workflow specializes the model for Najdi and Hijazi speech and mixes previously learned English and Arabic data into training to limit forgetting. A separate change that lets the model consider more future audio reduced word error rate by 1.31 absolute points without retraining, but added approximately 800 ms of latency. Nvidia limits its decoder conclusion to MALSD and says the workflow does not establish results for every Arabic dialect or deployment environment.

## Sources

- [NVIDIA Technical Blog](https://developer.nvidia.com/blog/fine-tuning-nvidia-nemotron-for-saudi-arabic-dialects-with-a-path-to-other-languages/)

---
Canonical: https://techandbusiness.org/newswire/D0S5lJzRMfdlMJ7au2bCkH
Published: 2026-10-01T06:06:47.170Z
Story chronology: 2026-10-01T05:00:00.000Z
Retrieved: 2026-10-01T08:19:38.685Z
Publisher: Tech & Business (techandbusiness.org)