arXiv：cs.CL· Ibrahim Almajai·· 3 小时前AI 评分31
Itgan 在 NADI 2026 共享任务：用 LoRA 参数高效微调 Whisper 实现鲁棒、混合方言与语码转换阿拉伯语 ASR
Itgan at NADI 2026 shared task: Parameter-Efficient Whisper Adaptation for Robust, Mixed-Dialect and Code-Switched Arabic ASR
AI 导读
Itgan 团队用消费级 GPU 上的 LoRA 微调 Whisper，一套方案覆盖 NADI 2026 三项阿拉伯语 ASR 子任务。鲁棒国家级方言任务取得 57.1% 国家平均 WER；混合方言任务为 46.7% WER；突尼斯语码转换任务以 14.49% WER 位列第二，CER 为 5.38%。最后 0.60 个 WER 点来自独立训练权重平均与 ROVER 投票，未再训练。
正文
Abstract: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. 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. A post-evaluation linear probe on frozen encoder features routes utterances without the label and recovers 44% of what oracle routing gives. On 1.2 the choice of base model mattered more than adapter capacity, and system combination helped only once we added a decorrelated member, reaching 46.7% WER. On 1.3 our system placed second at 14.49% WER with the lowest CER among the leading submissions, 5.38%. Its last 0.60 WER points came without further training, mostly from an exact weight-space average of independently trained runs, with ROVER voting adding the remainder. Every comparison carries a paired-bootstrap test, and we report eight directions that did not work.
| Comments: | 12 pages, Arabic NLP 2026 Shared Task |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Sound (cs.SD) |
| Cite as: | arXiv:2610.09934 [cs.CL] |
| (or arXiv:2610.09934v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.09934 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Ibrahim Almajai [view email]
[v1]
Wed, 7 Oct 2026 12:18:40 UTC (35 KB)
来源：arXiv：cs.CL · arxiv.org