LLM-Guided Reinforcement Learning for Audio-Visual Speech Enhancement

In existing Audio-Visual Speech Enhancement (AVSE) methods, objectives such as Scale-Invariant Signal-to-Noise Ratio (SI-SNR) and Mean Squared Error (MSE) are widely used; however, they often correlate poorly with perceptual quality and provide limited interpretability for optimization. This work proposes a reinforcement learning-based AVSE framework with a Large Language Model (LLM)-based interpretable reward model. An audio LLM generates natural language descriptions of enhanced speech, which are converted by a sentiment analysis model into a 1-5 rating score serving as the PPO reward for fine-tuning a pretrained AVSE model. Compared with scalar metrics, LLM-generated feedback is semantically rich and explicitly describes improvements in speech quality. Experiments on the 4th COG-MHEAR AVSE Challenge (AVSEC-4) dataset show that the proposed method outperforms a supervised baseline and a DNSMOS-based RL baseline in PESQ, STOI, neural quality metrics, and subjective listening tests.

Polyglot-Lion: Efficient Multilingual ASR for Singapore via Balanced Fine-Tuning of Qwen3-ASR

We present Polyglot-Lion, a family of compact multilingual automatic speech recognition (ASR) models tailored for the linguistic landscape of Singapore, covering English, Mandarin, Tamil, and Malay. Our models are obtained by fine-tuning Qwen3-ASR-0.6B and Qwen3-ASR-1.7B exclusively on publicly available speech corpora, using a balanced sampling strategy that equalizes the number of training utterances per language and deliberately omits language-tag conditioning so that the model learns to identify languages implicitly from audio. On 12 benchmarks spanning the four target languages, Polyglot-Lion-1.7B achieves an average error rate of 14.85, competitive with MERaLiON-2-10B-ASR (14.32) - a model 6x larger - while incurring a training cost of \$81 on a single RTX PRO 6000 GPU compared to \$18,862 for the 128-GPU baseline. Inference throughput is approximately 20x faster than MERaLiON at 0.10 s/sample versus 2.02 s/sample. These results demonstrate that linguistically balanced fine-tuning of moderate-scale pretrained models can yield deployment-ready multilingual ASR at a fraction of the cost of larger specialist systems.

Fanar 2.0: Arabic Generative AI Stack

We present Fanar 2.0, the second generation of Qatar's Arabic-centric Generative AI platform. Sovereignty is a first-class design principle: every component, from data pipelines to deployment infrastructure, was designed and operated entirely at QCRI, Hamad Bin Khalifa University. Fanar 2.0 is a story of resource-constrained excellence: the effort ran on 256 NVIDIA H100 GPUs, with Arabic having only ~0.5% of web data despite 400 million native speakers. Fanar 2.0 adopts a disciplined strategy of data quality over quantity, targeted continual pre-training, and model merging to achieve substantial gains within these constraints. At the core is Fanar-27B, continually pre-trained from a Gemma-3-27B backbone on a curated corpus of 120 billion high-quality tokens across three data recipes. Despite using 8x fewer pre-training tokens than Fanar 1.0, it delivers substantial benchmark improvements: Arabic knowledge (+9.1 pts), language (+7.3 pts), dialects (+3.5 pts), and English capability (+7.6 pts). Beyond the core LLM, Fanar 2.0 introduces a rich stack of new capabilities. FanarGuard is a state-of-the-art 4B bilingual moderation filter for Arabic safety and cultural alignment. The speech family Aura gains a long-form ASR model for hours-long audio. Oryx vision family adds Arabic-aware image and video understanding alongside culturally grounded image generation. An agentic tool-calling framework enables multi-step workflows. Fanar-Sadiq utilizes a multi-agent architecture for Islamic content. Fanar-Diwan provides classical Arabic poetry generation. FanarShaheen delivers LLM-powered bilingual translation. A redesigned multi-layer orchestrator coordinates all components through intent-aware routing and defense-in-depth safety validation. Taken together, Fanar 2.0 demonstrates that sovereign, resource-constrained AI development can produce systems competitive with those built at far greater scale.

HRTF-guided Binaural Target Speaker Extraction with Real-World Validation

This paper presents a Head-Related Transfer Function (HRTF)-guided framework for binaural Target Speaker Extraction (TSE) from mixtures of concurrent sources. Unlike conventional TSE methods based on Direction of Arrival (DOA) estimation or enrollment signals, which often distort perceived spatial location, the proposed approach leverages the listener's HRTF as an explicit spatial prior. The proposed framework is built upon a multi-channel deep blind source separation backbone, adapted to the binaural TSE setting. It is trained on measured HRTFs from a diverse population, enabling cross-listener generalization rather than subject-specific tuning. By conditioning the extraction on HRTF-derived spatial information, the method preserves binaural cues while enhancing speech quality and intelligibility. The performance of the proposed framework is validated through simulations and real recordings obtained from a head and torso simulator (HATS).

VorTEX: Various overlap ratio for Target speech EXtraction

Target speech extraction (TSE) aims to recover a target speaker's voice from a mixture. While recent text-prompted approaches have shown promise, most approaches assume fully overlapped mixtures, limiting insight into behavior across realistic overlap ratios. We introduce VorTEX (Various overlap ratio for Target speech EXtraction), a text-prompted TSE architecture with a Decoupled Adaptive Multi-branch (DAM) Fusion block that separates primary extraction from auxiliary regularization pathways. To enable controlled analysis, we construct PORTE, a two-speaker dataset spanning overlap ratios from 0% to 100%. We further propose Suppression Ratio on Energy (SuRE), a diagnostic metric that detects suppression behavior not captured by conventional measures. Experiments show that existing models exhibit suppression or residual interference under overlap, whereas VorTEX achieves the highest separation fidelity across 20-100% overlap (e.g., 5.50 dB at 20% and 2.04 dB at 100%) while maintaining zero SuRE, indicating robust extraction without suppression-driven artifacts.

Speak, Segment, Track, Navigate: An Interactive System for Video-Guided Skull-Base Surgery

We introduce a speech-guided embodied agent framework for video-guided skull base surgery that dynamically executes perception and image-guidance tasks in response to surgeon queries. The proposed system integrates natural language interaction with real-time visual perception directly on live intraoperative video streams, thereby enabling surgeons to request computational assistance without disengaging from operative tasks. Unlike conventional image-guided navigation systems that rely on external optical trackers and additional hardware setup, the framework operates purely on intraoperative video. The system begins with interactive segmentation and labeling of the surgical instrument. The segmented instrument is then used as a spatial anchor that is autonomously tracked in the video stream to support downstream workflows, including anatomical segmentation, interactive registration of preoperative 3D models, monocular video-based estimation of the surgical tool pose, and support image guidance through real-time anatomical overlays.We evaluate the proposed system in video-guided skull base surgery scenarios and benchmark its tracking performance against a commercially available optical tracking system. Results demonstrate that speech-guided embodied agents can achieve competitive spatial accuracy while improving workflow integration and enabling rapid deployment of video-guided surgical systems.

CAST-TTS: A Simple Cross-Attention Framework for Unified Timbre Control in TTS

Current Text-to-Speech (TTS) systems typically use separate models for speech-prompted and text-prompted timbre control. While unifying both control signals into a single model is desirable, the challenge of cross-modal alignment often results in overly complex architectures and training objective. To address this challenge, we propose CAST-TTS, a simple yet effective framework for unified timbre control. Features are extracted from speech prompts and text prompts using pre-trained encoders. The multi-stage training strategy efficiently aligns the speech and projected text representations within a shared embedding space. A single cross-attention mechanism then allows the model to use either of these representations to control the timbre. Extensive experiments validate that the unified cross-attention mechanism is critical for achieving high-quality synthesis. CAST-TTS achieves performance comparable to specialized single-input models while operating within a unified architecture. The demo page can be accessed at https://HiRookie9.github.io/CAST-TTS-Page.

Attention-guided Evidence Grounding for Spoken Question Answering

Spoken Question Answering (Spoken QA) presents a challenging cross-modal problem: effectively aligning acoustic queries with textual knowledge while avoiding the latency and error propagation inherent in cascaded ASR-based systems. In this paper, we introduce Attention-guided Evidence Grounding (AEG), a novel end-to-end framework that leverages the internal cross-modal attention of Speech Large Language Models (SpeechLLMs) to explicitly locate and ground key evidence in the model's latent space. To address the diffuse attention distribution in pre-trained models, we propose Learning to Focus on Evidence (LFE), a supervised fine-tuning paradigm that calibrates the model's attention mechanism to distinguish query-relevant segments from irrelevant context. Experiments on SQuAD, HotpotQA, and MuSiQue demonstrate that AEG reduces hallucinations and achieves strong efficiency gains, outperforming large-scale cascaded baselines (Whisper-Large-v3 + Reranker) while reducing inference latency by approximately 62%.

Is Semi-Automatic Transcription Useful in Corpus Creation? Preliminary Considerations on the KIParla Corpus

This paper analyses the implementation of Automatic Speech Recognition (ASR) into the transcription workflow of the KIParla corpus, a resource of spoken Italian. Through a two-phase experiment, 11 expert and novice transcribers produced both manual and ASR-assisted transcriptions of identical audio segments across three different types of conversation, which were subsequently analyzed through a combination of statistical modeling, word-level alignment and a series of annotation-based metrics. Results show that ASR-assisted workflows can increase transcription speed but do not consistently improve overall accuracy, with effects depending on multiple factors such as workflow configuration, conversation type and annotator experience. Analyses combining alignment-based metrics, descriptive statistics and statistical modeling provide a systematic framework to monitor transcription behavior across annotators and workflows. Despite limitations, ASR-assisted transcription, potentially supported by task-specific fine-tuning, could be integrated into the KIParla transcription workflow to accelerate corpus creation without compromising transcription quality.

SEAHateCheck: Functional Tests for Detecting Hate Speech in Low-Resource Languages of Southeast Asia

Hate speech detection relies heavily on linguistic resources, which are primarily available in high-resource languages such as English and Chinese, creating barriers for researchers and platforms developing tools for low-resource languages in Southeast Asia, where diverse socio-linguistic contexts complicate online hate moderation. To address this, we introduce SEAHateCheck, a pioneering dataset tailored to Indonesia, Thailand, the Philippines, and Vietnam, covering Indonesian, Tagalog, Thai, and Vietnamese. Building on HateCheck's functional testing framework and refining SGHateCheck's methods, SEAHateCheck provides culturally relevant test cases, augmented by large language models and validated by local experts for accuracy. Experiments with state-of-the-art and multilingual models revealed limitations in detecting hate speech in specific low-resource languages. In particular, Tagalog test cases showed the lowest model accuracy, likely due to linguistic complexity and limited training data. In contrast, slang-based functional tests proved the hardest, as models struggled with culturally nuanced expressions. The diagnostic insights of SEAHateCheck further exposed model weaknesses in implicit hate detection and models' struggles with counter-speech expression. As the first functional test suite for these Southeast Asian languages, this work equips researchers with a robust benchmark, advancing the development of practical, culturally attuned hate speech detection tools for inclusive online content moderation.