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.
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.
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.
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.
The paper presents the Itgan systems for the NADI 2026 Arabic ASR shared task, covering robust country-level, mixed-dialect, and Tunisian code‑switched subtasks. All systems use a Whisper model adapted with LoRA on consumer GPUs, with task‑specific additions: per‑dialect specialists for 1.1, base‑model choice and decorrelated ensemble for 1.2, and a weight‑space average plus ROVER voting for 1.3. The submissions achieved 57.1% WER on 1.1, 46.7% WER on 1.2, and 14.49% WER on 1.3, with detailed statistical comparisons and noted ineffective approaches.
ufakzeka-kulak is an open Turkish speech‑to‑text model with 31.6 M parameters, built using a FastConformer encoder, a token‑and‑duration transducer (TDT), and a CTC head. It was trained from scratch on 4,977 hours of data—281 hours of human transcription and 4,696 hours of machine‑generated subtitles that were verified by a second reading—under licenses permitting commercial use. The model outputs cased, punctuated Turkish text with numbers written as words, achieving a normalised WER of 0.0921 on FLEURS Turkish and 0.0916 on Common Voice 27.0 Turkish, outperforming Whisper base and small models, and it is released under Apache‑2.0.
O'ido is an inference engine that performs open‑vocabulary English speech recognition entirely on an ESP32‑S3 microcontroller, running a 13‑million‑parameter Conformer‑CTC model with int8 quantization and a 1.3‑million‑parameter language model. On LibriSpeech it achieves 3.70%/8.23% WER on test‑clean/test‑other, only 0.12 points below full precision, and outperforms the vendor’s recognizer and several laptop‑based tiny models. The firmware processes audio in about twice the duration of the input on a single core, with a streaming variant that delivers partial transcripts while speaking, though it is not yet real‑time. whyItMatters":"This work demonstrates that sophisticated, open‑vocabulary speech recognition can run on inexpensive, low‑power microcontrollers without dedicated neural accelerators, expanding the reach of voice interfaces to devices that previously could only support keyword spotting or fixed command sets."
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.
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.
The study investigates how to adapt automatic speech recognition (ASR) models to child speech while preserving adult performance, focusing on Arabic and English data. It compares full fine‑tuning, LoRA, and post‑hoc weight‑space merging across encoder–decoder, encoder–CTC, and AudioLLM‑based ASR systems, using child and adult benchmarks. Results indicate that child adaptation is essential, especially for non‑native speakers, but direct adaptation can degrade adult accuracy; bilingual adaptation and weight‑space merging techniques help balance the trade‑off, with encoder–decoder fine‑tuning still yielding the best raw WER for bilingual models.
arXiv:2609.18533v1 Announce Type: new Abstract: Automatic speech recognition (ASR) systems exhibit unequal error rates across speaker groups, motivating interventions on their internal representation...
The paper introduces the Offline AI Modules workstream, which provides a voice‑first offline architecture, a low‑cost hardware reference bill of materials, and a reproducible quantization and benchmarking pipeline for instruction‑tuned language models in the 2‑5B parameter range. It evaluates three models across four quantization formats on two hardware tiers—NVIDIA Jetson Orin NX and Raspberry Pi5—measuring deployment metrics and multilingual quality. The key result is that Q4_K_M quantization offers the best size‑to‑quality trade‑off, enabling high decode throughput and strong topic classification accuracy while staying within memory limits on both tiers.
arXiv:2609. 11786v1 Announce Type: new Abstract: Automatic speech recognition (ASR) systems and audio language models (audio LMs) now report low error rates on monolingual benchmarks, but their behavior on code switched speech in low resource, diacritic rich languages remains poorly characterized.