D-ORCA: Dialogue-Centric Optimization for Robust Audio-Visual Captioning

Spoken dialogue is a primary source of information in videos; therefore, accurately identifying who spoke what and when is essential for deep video understanding. We introduce D-ORCA, a \textbf{d}ialogue-centric \textbf{o}mni-modal large language model optimized for \textbf{r}obust audio-visual \textbf{ca}ptioning. We further curate DVD, a large-scale, high-quality bilingual dataset comprising nearly 40,000 multi-party dialogue videos for training and 2000 videos for evaluation in English and Mandarin, addressing a critical gap in the open-source ecosystem. To ensure fine-grained captioning accuracy, we adopt group relative policy optimization with three novel reward functions that assess speaker attribution accuracy, global speech content accuracy, and sentence-level temporal boundary alignment. These rewards are derived from evaluation metrics widely used in speech processing and, to our knowledge, are applied for the first time as reinforcement learning objectives for audio-visual captioning. Extensive experiments demonstrate that D-ORCA substantially outperforms existing open-source models in speaker identification, speech recognition, and temporal grounding. Notably, despite having only 8 billion parameters, D-ORCA achieves performance competitive with Qwen3-Omni across several general-purpose audio-visual understanding benchmarks. Demos are available at \href{https://d-orca-llm.github.io/}{https://d-orca-llm.github.io/}. Our code, data, and checkpoints will be available at \href{https://github.com/WeChatCV/D-ORCA/}{https://github.com/WeChatCV/D-ORCA/}.

Adapting Where It Matters: Depth-Aware Adaptation for Efficient Multilingual Speech Recognition in Low-Resource Languages

Recent speech foundation models excel at multilingual automatic speech recognition (ASR) for high-resource languages, but adapting them to low-resource languages remains challenging due to data scarcity and efficiency constraints. Full-model fine-tuning is computationally expensive and prone to overfitting, while parameter-efficient methods like LoRA apply adaptation uniformly across layers, overlooking internal representations thus compromising effectiveness and efficiency. We analyze multilingual ASR models and reveal a U-shaped adaptability pattern: early and late layers are language-specific and require more adaptation, while intermediate layers retain shared semantics and need less. Building on this observation, we propose DAMA, a Depth-Aware Model Adaptation framework that allocates adaptation capacity according to each layer's role. DAMA also introduces Singular Value Decomposition (SVD)-based initialization to constrain adaptation and preserve the U-shaped pattern, as well as a frozen middle-layer basis for further efficiency. Evaluated on 18 low-resource languages across two benchmark datasets, DAMA matches or surpasses state-of-the-art accuracy with 80% fewer trainable parameters, achieves a 29% error reduction under extreme data scarcity, and significantly improves memory, training time, and computational efficiency over baselines. These results highlight the benefits of structure-aware adaptation for efficient, scalable multilingual ASR.

From Hallucination to Articulation: Language Model-Driven Losses for Ultra Low-Bitrate Neural Speech Coding

``Phoneme Hallucinations (PH)'' commonly occur in low-bitrate DNN-based codecs. It is the generative decoder's attempt to synthesize plausible outputs from excessively compressed tokens missing some semantic information. In this work, we propose language model-driven losses (LM loss) and show they may alleviate PHs better than a semantic distillation (SD) objective in very-low-bitrate settings. The proposed LM losses build upon language models pretrained to associate speech with text. When ground-truth transcripts are unavailable, we propose to modify a popular automatic speech recognition (ASR) model, Whisper, to compare the decoded utterance against the ASR-inferred transcriptions of the input speech. Else, we propose to use the timed-text regularizer (TTR) to compare WavLM representations of the decoded utterance against BERT representations of the ground-truth transcriptions. We test and compare LM losses against an SD objective, using a reference codec whose three-stage training regimen was designed after several popular codecs. Subjective and objective evaluations conclude that LM losses may provide stronger guidance to extract semantic information from self-supervised speech representations, boosting human-perceived semantic adherence while preserving overall output quality. Demo samples, code, and checkpoints are available online.

EmoAra: Emotion-Preserving English Speech Transcription and Cross-Lingual Translation with Arabic Text-to-Speech

This work presents EmoAra, an end-to-end emotion-preserving pipeline for cross-lingual spoken communication, motivated by banking customer service where emotional context affects service quality. EmoAra integrates Speech Emotion Recognition, Automatic Speech Recognition, Machine Translation, and Text-to-Speech to process English speech and deliver an Arabic spoken output while retaining emotional nuance. The system uses a CNN-based emotion classifier, Whisper for English transcription, a fine-tuned MarianMT model for English-to-Arabic translation, and MMS-TTS-Ara for Arabic speech synthesis. Experiments report an F1-score of 94% for emotion classification, translation performance of BLEU 56 and BERTScore F1 88.7%, and an average human evaluation score of 81% on banking-domain translations. The implementation and resources are available at the accompanying GitHub repository.

Benchmarking Automatic Speech Recognition for Indian Languages in Agricultural Contexts

The digitization of agricultural advisory services in India requires robust Automatic Speech Recognition (ASR) systems capable of accurately transcribing domain-specific terminology in multiple Indian languages. This paper presents a benchmarking framework for evaluating ASR performance in agricultural contexts across Hindi, Telugu, and Odia languages. We introduce evaluation metrics including Agriculture Weighted Word Error Rate (AWWER) and domain-specific utility scoring to complement traditional metrics. Our evaluation of 10,934 audio recordings, each transcribed by up to 10 ASR models, reveals performance variations across languages and models, with Hindi achieving the best overall performance (WER: 16.2%) while Odia presents the greatest challenges (best WER: 35.1%, achieved only with speaker diarization). We characterize audio quality challenges inherent to real-world agricultural field recordings and demonstrate that speaker diarization with best-speaker selection can substantially reduce WER for multi-speaker recordings (upto 66% depending on the proportion of multi-speaker audio). We identify recurring error patterns in agricultural terminology and provide practical recommendations for improving ASR systems in low-resource agricultural domains. The study establishes baseline benchmarks for future agricultural ASR development.

Attention-weighted Centered Kernel Alignment for Knowledge Distillation in Large Audio-Language Models Applied to Speech Emotion Recognition

The emergence of Large Audio-Language Models (LALMs) has advanced Speech Emotion Recognition (SER), but their size limits deployment in resource-constrained environments. While Knowledge Distillation is effective for LALM compression, existing methods remain underexplored in distilling the cross-modal projection module (Projector), and often struggle with alignment due to differences in feature dimensions. We propose PL-Distill, a KD framework that combines Projector-Level Distillation (PDist) to align audio embeddings and Logits-Level Distillation (LDist) to align output logits. PDist introduces Attention-weighted Centered Kernel Alignment, a novel approach we propose to highlight important time steps and address dimension mismatches. Meanwhile, LDist minimizes the Kullback-Leibler divergence between teacher and student logits from audio and text modalities. On IEMOCAP, RAVDESS, and SAVEE, PL-Distill compresses an 8.4B-parameter teacher to a compact 1.1B-parameter student, consistently outperforming the teacher, state-of-the-art pretrained models, and other KD baselines across all metrics.

asr_eval: Algorithms and tools for multi-reference and streaming speech recognition evaluation

We propose several improvements to the speech recognition evaluation. First, we propose a string alignment algorithm that supports both multi-reference labeling, arbitrary-length insertions and better word alignment. This is especially useful for non-Latin languages, those with rich word formation, to label cluttered or longform speech. Secondly, we collect a novel test set DiverseSpeech-Ru of longform in-the-wild Russian speech with careful multi-reference labeling. We also perform multi-reference relabeling of popular Russian tests set and study fine-tuning dynamics on its corresponding train set. We demonstrate that the model often adopts to dataset-specific labeling, causing an illusion of metric improvement. Based on the improved word alignment, we develop tools to evaluate streaming speech recognition and to align multiple transcriptions to compare them visually. Additionally, we provide uniform wrappers for many offline and streaming speech recognition models. Our code will be made publicly available.

Streaming Speech Recognition with Decoder-Only Large Language Models and Latency Optimization

Recent advances have demonstrated the potential of decoderonly large language models (LLMs) for automatic speech recognition (ASR). However, enabling streaming recognition within this framework remains a challenge. In this work, we propose a novel streaming ASR approach that integrates a read/write policy network with monotonic chunkwise attention (MoChA) to dynamically segment speech embeddings. These segments are interleaved with label sequences during training, enabling seamless integration with the LLM. During inference, the audio stream is buffered until the MoChA module triggers a read signal, at which point the buffered segment together with the previous token is fed into the LLM for the next token prediction. We also introduce a minimal-latency training objective to guide the policy network toward accurate segmentation boundaries. Furthermore, we adopt a joint training strategy in which a non-streaming LLM-ASR model and our streaming model share parameters. Experiments on the AISHELL-1 and AISHELL-2 Mandarin benchmarks demonstrate that our method consistently outperforms recent streaming ASR baselines, achieving character error rates of 5.1% and 5.5%, respectively. The latency optimization results in a 62.5% reduction in average token generation delay with negligible impact on recognition accuracy

VowelPrompt: Hearing Speech Emotions from Text via Vowel-level Prosodic Augmentation

Emotion recognition in speech presents a complex multimodal challenge, requiring comprehension of both linguistic content and vocal expressivity, particularly prosodic features such as fundamental frequency, intensity, and temporal dynamics. Although large language models (LLMs) have shown promise in reasoning over textual transcriptions for emotion recognition, they typically neglect fine-grained prosodic information, limiting their effectiveness and interpretability. In this work, we propose VowelPrompt, a linguistically grounded framework that augments LLM-based emotion recognition with interpretable, fine-grained vowel-level prosodic cues. Drawing on phonetic evidence that vowels serve as primary carriers of affective prosody, VowelPrompt extracts pitch-, energy-, and duration-based descriptors from time-aligned vowel segments, and converts these features into natural language descriptions for better interpretability. Such a design enables LLMs to jointly reason over lexical semantics and fine-grained prosodic variation. Moreover, we adopt a two-stage adaptation procedure comprising supervised fine-tuning (SFT) followed by Reinforcement Learning with Verifiable Reward (RLVR), implemented via Group Relative Policy Optimization (GRPO), to enhance reasoning capability, enforce structured output adherence, and improve generalization across domains and speaker variations. Extensive evaluations across diverse benchmark datasets demonstrate that VowelPrompt consistently outperforms state-of-the-art emotion recognition methods under zero-shot, fine-tuned, cross-domain, and cross-linguistic conditions, while enabling the generation of interpretable explanations that are jointly grounded in contextual semantics and fine-grained prosodic structure.

Towards Robust Dysarthric Speech Recognition: LLM-Agent Post-ASR Correction Beyond WER

While Automatic Speech Recognition (ASR) is typically benchmarked by word error rate (WER), real-world applications ultimately hinge on semantic fidelity. This mismatch is particularly problematic for dysarthric speech, where articulatory imprecision and disfluencies can cause severe semantic distortions. To bridge this gap, we introduce a Large Language Model (LLM)-based agent for post-ASR correction: a Judge-Editor over the top-k ASR hypotheses that keeps high-confidence spans, rewrites uncertain segments, and operates in both zero-shot and fine-tuned modes. In parallel, we release SAP-Hypo5, the largest benchmark for dysarthric speech correction, to enable reproducibility and future exploration. Under multi-perspective evaluation, our agent achieves a 14.51% WER reduction alongside substantial semantic gains, including a +7.59 pp improvement in MENLI and +7.66 pp in Slot Micro F1 on challenging samples. Our analysis further reveals that WER is highly sensitive to domain shift, whereas semantic metrics correlate more closely with downstream task performance.