Itgan at NADI 2026 shared task: Parameter-Efficient Whisper Adaptation for Robust, Mixed-Dialect and Code-Switched Arabic ASR

Key points

• All three share one recipe, Whisper adapted with LoRA on consumer GPUs, and each was carried by a different addition to it.1
• On 1.1, where the dialect label is given at test time, per-dialect specialists continued from a pooled adapter gave the largest gain, and the submitted system reached 57.1% country-average WER.1
• A post-evaluation linear probe on frozen encoder features routes utterances without the label and recovers 44% of what oracle routing gives.1
• On 1.3 our system placed second at 14.49% WER with the lowest CER among the leading submissions, 5.38%.1

Sources (1)

• [1]Itgan at NADI 2026 shared task: Parameter-Efficient Whisper Adaptation for Robust, Mixed-Dialect and Code-Switched Arabic ASR
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 7, 12:18 PM

  We describe the Itgan systems for the three ASR subtasks of NADI 2026, namely robust country-level ASR (1.1), mixed-dialect ASR (1.2), and Tunisian code-switched ASR (1.3).

  All three share one recipe, Whisper adapted with LoRA on consumer GPUs, and each was carried by a different addition to it.

Extractive summary: sentences quoted from the sources.