The paper evaluates a 305 M‑parameter audio encoder shared by Gemma 4 and EmbeddingGemma 2 for Saudi and Gulf Arabic speech recognition without altering its weights. Two approaches are compared: a frozen system that trains only a 22.6 M‑parameter CTC reader, and an adapted system that adds 3.1 M low‑rank adapters, a 41.5 M‑parameter hybrid CTC/attention reader, and joint decoding with a sub‑word language model. The adapted system matches or surpasses a 0.6 B‑parameter fine‑tuned recogniser on several datasets, while the frozen system remains competitive in character error rate and runs three‑quarters faster on CPU with greedy decoding.
By Amr Abdelkhalik
The paper introduces a new benchmark of 13,044 spontaneous Nepali‑English code‑switched podcast utterances, totaling 15.45 h, and evaluates a two‑stage synthetic‑to‑real adaptation framework for ASR. Stage 1 uses supervised fine‑tuning on 0.95 h of synthetic speech, reducing Mixed Error Rate from 23.40 % to 21.52 % and switch‑point error rate from 35.89 % to 28.77 %. Stage 2 applies Group Relative Policy Optimization with script and transcription rewards, yielding modest but statistically significant further improvements, particularly in English script confusion and switch‑point accuracy. The study also shows that an auxiliary script reward term offers no benefit beyond MER‑only optimization in this setting, and that the synthetic SFT provides the main cross‑domain transfer to spontaneous human speech.
By Firoj Paudel
arXiv:2608. 12327v1 Announce Type: cross Abstract: Multilingual pretrained models nominally support Nepali, yet no controlled benchmark has compared them under a single fine-tuning protocol.
By Suman Paudel, Sarbin Sayami
arXiv:2606. 24169v1 Announce Type: new Abstract: Adapting a streaming speech recognition model to a new language requires choosing between two plausible warm starts: a multilingual (ML) encoder or an English-only (EN) encoder.
By Nenad Banfic
The paper introduces WaxalNet, a benchmark and adaptation study for spontaneous speech recognition across 19 African languages using the WAXAL corpus. It evaluates zero‑shot and fine‑tuned performance of several multilingual foundation models (Whisper, MMS, Omnilingual ASR, SeamlessM4T‑v2), showing that fine‑tuning can substantially reduce word error rates and that LLM decoders generally outperform CTC decoders when the encoder is shared. The authors also demonstrate that a single multilingual model can serve all languages with only a modest WER penalty and provide a native‑speaker audit of failure modes.
By Victor Tolulope Olufemi, Oreoluwa Babatunde, Ramsey Njema, Bolarinwa Gbotemi, Wanchi Lucia Yen, John Uzodinma, Sunday Ajayi, Oluwademilade Williams, Kausar Moshood, Innocent Elendu Anyaele, Akebert Arefaine, Candace Hunzwi, Wongel Dawit Daniel, Emmilly Namuganga, Cleophas Kadima, Athanase Bahizire, Onitsiky Ranaivoson, Emmanuel Aaron, Nicholaus Ladislaus, Idris Muhammed, Jonathan Enoch Simenya, Martin Koome, Matewos Tegete Endaylalu, Peter Ifeoluwa Adeyemo, Hondi Prisca Birindwa, Ukachi Agnes Eze-Mbey, Yacoba Oduro-Yeboah, Pericles Adjovi, Mikel K. Ngueajio, Toluwani Aremu, Prasenjit Mitra
arXiv:2606. 07608v1 Announce Type: cross Abstract: We present a systematic study of fine-tuning OpenAI's Whisper large-v3 for Swiss German ASR, using 1,367 hours of broadcast speech paired with Standard German subtitles as weak supervision.
By Felix Akeret
AfriSwitch is a 61.36‑hour, human‑transcribed benchmark of in‑the‑wild code‑switched speech covering 16 African languages and varieties, annotated with switch‑level English span tags, per‑utterance Code‑Mixing Index (CMI), and switch‑point counts. The corpus reveals that code‑switching behaviour varies widely across languages, with no single metric fully capturing how code‑switched a language is. Benchmarking five open and commercial multilingual ASR systems in a zero‑shot setting shows high word error rates, with the best system averaging 35.93% WER and none dropping below 24% on any language, indicating that Africa‑targeted training rather than model scale or nominal language coverage best predicts performance.
By Gabrial Zencha Ashungafac, Busayo Awobade, Tobi Olatunji
arXiv:2606. 17255v1 Announce Type: cross Abstract: This work describes the participation of the MLLP-VRAIN research group in the shared task of the IWSLT 2026 Simultaneous Speech Translation track.
By Jorge Iranzo-S\'anchez, Gerard Mas-Moll\`a, Adri\`a Gim\'enez, Jorge Civera, Albert Sanchis, Alfons Juan
The paper evaluates a multilingual ASR model (MMS‑1B‑all) on a Garrusi Kurdish dataset using a common‑reference staged normalization approach. By normalizing both reference and hypothesis, the authors show that raw Arabic‑script hypotheses yield a 111.70 % WER, which drops to 97.85 % after folding into a reduced orthography, highlighting the impact of orthographic differences on error measurement. A Southern Kurdish fine‑tuned system performs worse, and residual errors are partly due to scoring‑pipeline limitations rather than recognition failures.
By Hiwa Asadpour
NADI 2026 is the seventh edition of the Nuanced Arabic Dialect Identification shared task series and the second focused on multidialectal Arabic speech processing. It includes five main tasks—Automatic Speech Recognition, Spoken Dialect Identification, Text-to-Speech, Spoken Language Translation, and Spoken Language Understanding—along with eight subtasks that test realistic scenarios such as low‑bandwidth, mixed dialects, code‑switching, out‑of‑domain, and zero‑shot settings. The event attracted 21 teams from at least 13 countries, with 48 test‑phase submissions and 14 system‑description papers, and the results highlight out‑of‑domain generalization as a major bottleneck while showcasing the strengths of Arabic‑specialized speech models, multimodal dialect identification, and ensemble methods.
By Peter Sullivan, Bashar Talafha, Ahmed Ashraf, Fethi Bougares, Haroun Elleuch, Chiyu Zhang, AbdelRahim Elmadany, Youssef Mohamed, Salima Mdhaffar, Yannick Est\`eve, Mohamed Elhoseiny, Hamzah Luqman, Nizar Habash, Muhammad Abdul-Mageed
The paper introduces a method to prune six layers from the encoder of OpenAI’s Whisper ASR model, reducing the encoder stack by 18.5% without requiring custom inference code. Layers are selected based on their minimal impact on Word Error Rate when removed. After pruning, the model’s WER rises from 18.2% to 21.9%, but distillation with unlabeled monolingual speech data lowers it to 20.1%.
"whyItMatters":"The approach offers a straightforward way to accelerate Whisper inference by simplifying the encoder while maintaining acceptable accuracy, and the released code and model enable immediate adoption by the community."
By Rasmus Aagaard, Nicki Skafte Detlefsen
arXiv:2610.20263v1 Announce Type: new
Abstract: Every multilingual speech pipeline begins with a language decision, an error there propagates silently through recognition, translation and routing. In...
By Stephen Fernandes, Narayan Choudhary