Distilled HuBERT for Mobile Speech Emotion Recognition: A Cross-Corpus Validation Study

Speech Emotion Recognition (SER) has significant potential for mobile applications, yet deployment remains constrained by the computational demands of state-of-the-art transformer architectures. This paper presents a mobile-efficient SER system based on DistilHuBERT, a distilled and 8-bit quantized transformer that achieves approximately 92% parameter reduction compared to full-scale Wav2Vec 2.0 models while maintaining competitive accuracy. We conduct a rigorous 5-fold Leave-One-Session-Out (LOSO) cross-validation on the IEMOCAP dataset to ensure speaker independence, augmented with cross-corpus training on CREMA-D to enhance generalization. Cross-corpus training with CREMA-D yields a 1.2% improvement in Weighted Accuracy, a 1.4% gain in Macro F1-score, and a 32% reduction in cross-fold variance, with the Neutral class showing the most substantial benefit at 5.4% F1-score improvement. Our approach achieves an Unweighted Accuracy of 61.4% with a quantized model footprint of only 23 MB, representing approximately 91% of the Unweighted Accuracy of a full-scale baseline. Cross-corpus evaluation on RAVDESS reveals that the theatrical nature of acted emotions causes predictions to cluster by arousal level rather than by specific emotion categories - happiness predictions systematically bleed into anger predictions, and sadness predictions bleed into neutral predictions, due to acoustic saturation when actors prioritize clarity over subtlety. Despite this theatricality effect reducing overall RAVDESS accuracy to 46.64%, the model maintains robust arousal detection with 99% recall for anger, 55% recall for neutral, and 27% recall for sadness. These findings demonstrate a Pareto-optimal tradeoff between model size and accuracy, enabling practical affect recognition on resource-constrained mobile devices.

VALLR-Pin: Uncertainty-Factorized Visual Speech Recognition for Mandarin with Pinyin Guidance

Visual speech recognition (VSR) aims to transcribe spoken content from silent lip-motion videos and is particularly challenging in Mandarin due to severe viseme ambiguity and pervasive homophones. We propose VALLR-Pin, a two-stage Mandarin VSR framework that extends the VALLR architecture by explicitly incorporating Pinyin as an intermediate representation. In the first stage, a shared visual encoder feeds dual decoders that jointly predict Mandarin characters and their corresponding Pinyin sequences, encouraging more robust visual-linguistic representations. In the second stage, an LLM-based refinement module takes the predicted Pinyin sequence together with an N-best list of character hypotheses to resolve homophone-induced ambiguities. To further adapt the LLM to visual recognition errors, we fine-tune it on synthetic instruction data constructed from model-generated Pinyin-text pairs, enabling error-aware correction. Experiments on public Mandarin VSR benchmarks demonstrate that VALLR-Pin consistently improves transcription accuracy under multi-speaker conditions, highlighting the effectiveness of combining phonetic guidance with lightweight LLM refinement.

PROFASR-BENCH: A Benchmark for Context-Conditioned ASR in High-Stakes Professional Speech

Automatic Speech Recognition (ASR) in professional settings faces challenges that existing benchmarks underplay: dense domain terminology, formal register variation, and near-zero tolerance for critical entity errors. We present ProfASR-Bench, a professional-talk evaluation suite for high-stakes applications across finance, medicine, legal, and technology. Each example pairs a natural-language prompt (domain cue and/or speaker profile) with an entity-rich target utterance, enabling controlled measurement of context-conditioned recognition. The corpus supports conventional ASR metrics alongside entity-aware scores and slice-wise reporting by accent and gender. Using representative families Whisper (encoder-decoder ASR) and Qwen-Omni (audio language models) under matched no-context, profile, domain+profile, oracle, and adversarial conditions, we find a consistent pattern: lightweight textual context produces little to no change in average word error rate (WER), even with oracle prompts, and adversarial prompts do not reliably degrade performance. We term this the context-utilization gap (CUG): current systems are nominally promptable yet underuse readily available side information. ProfASR-Bench provides a standardized context ladder, entity- and slice-aware reporting with confidence intervals, and a reproducible testbed for comparing fusion strategies across model families. Dataset: https://huggingface.co/datasets/prdeepakbabu/ProfASR-Bench Code: https://github.com/prdeepakbabu/ProfASR-Bench

Contextual Biasing for LLM-Based ASR with Hotword Retrieval and Reinforcement Learning

Large language model (LLM)-based automatic speech recognition (ASR) has recently achieved strong performance across diverse tasks, yet contextual biasing for named entities and hotwords under large vocabularies remains challenging. In this work, we propose a scalable two-stage framework that integrates hotword retrieval with LLM-ASR adaptation. First, we extend the Global-Local Contrastive Language-Audio pre-trained model (GLCLAP) to retrieve a compact top-k set of hotword candidates from a large vocabulary via robustness-aware data augmentation and fuzzy matching. Second, we inject the retrieved candidates as textual prompts into an LLM-ASR model and fine-tune it with Generative Rejection-Based Policy Optimization (GRPO), using a task-driven reward that jointly optimizes hotword recognition and overall transcription accuracy. Experiments on hotword-focused test sets show substantial keyword error rate (KER) reductions while maintaining sentence accuracy on general ASR benchmarks, demonstrating the effectiveness of the proposed framework for large-vocabulary contextual biasing.

Rare Word Recognition and Translation Without Fine-Tuning via Task Vector in Speech Models

Rare words remain a critical bottleneck for speech-to-text systems. While direct fine-tuning improves recognition of target words, it often incurs high cost, catastrophic forgetting, and limited scalability. To address these challenges, we propose a training-free paradigm based on task vectors for rare word recognition and translation. By defining task vectors as parameter differences and introducing word-level task vector arithmetic, our approach enables flexible composition of rare-word capabilities, greatly enhancing scalability and reusability. Extensive experiments across multiple domains show that the proposed method matches or surpasses fine-tuned models on target words, improves general performance by about 5 BLEU, and mitigates catastrophic forgetting.

Semantic Codebooks as Effective Priors for Neural Speech Compression

Speech codecs are traditionally optimized for waveform fidelity, allocating bits to preserve acoustic detail even when much of it can be inferred from linguistic structure. This leads to inefficient compression and suboptimal performance on downstream recognition tasks. We propose SemDAC, a semantic-aware neural audio codec that leverages semantic codebooks as effective priors for speech compression. In SemDAC, the first quantizer in a residual vector quantization (RVQ) stack is distilled from HuBERT features to produce semantic tokens that capture phonetic content, while subsequent quantizers model residual acoustics. A FiLM-conditioned decoder reconstructs audio conditioned on the semantic tokens, improving efficiency in the use of acoustic codebooks. Despite its simplicity, this design proves highly effective: SemDAC outperforms DAC across perceptual metrics and achieves lower WER when running Whisper on reconstructed speech, all while operating at substantially lower bitrates (e.g., 0.95 kbps vs. 2.5 kbps for DAC). These results demonstrate that semantic codebooks provide an effective inductive bias for neural speech compression, producing compact yet recognition-friendly representations.

ElfCore: A 28nm Neural Processor Enabling Dynamic Structured Sparse Training and Online Self-Supervised Learning with Activity-Dependent Weight Update

In this paper, we present ElfCore, a 28nm digital spiking neural network processor tailored for event-driven sensory signal processing. ElfCore is the first to efficiently integrate: (1) a local online self-supervised learning engine that enables multi-layer temporal learning without labeled inputs; (2) a dynamic structured sparse training engine that supports high-accuracy sparse-to-sparse learning; and (3) an activity-dependent sparse weight update mechanism that selectively updates weights based solely on input activity and network dynamics. Demonstrated on tasks including gesture recognition, speech, and biomedical signal processing, ElfCore outperforms state-of-the-art solutions with up to 16X lower power consumption, 3.8X reduced on-chip memory requirements, and 5.9X greater network capacity efficiency.

VALLR-Pin: Dual-Decoding Visual Speech Recognition for Mandarin with Pinyin-Guided LLM Refinement

Visual Speech Recognition aims to transcribe spoken words from silent lip-motion videos. This task is particularly challenging for Mandarin, as visemes are highly ambiguous and homophones are prevalent. We propose VALLR-Pin, a novel two-stage framework that extends the recent VALLR architecture from English to Mandarin. First, a shared video encoder feeds into dual decoders, which jointly predict both Chinese character sequences and their standard Pinyin romanization. The multi-task learning of character and phonetic outputs fosters robust visual-semantic representations. During inference, the text decoder generates multiple candidate transcripts. We construct a prompt by concatenating the Pinyin output with these candidate Chinese sequences and feed it to a large language model to resolve ambiguities and refine the transcription. This provides the LLM with explicit phonetic context to correct homophone-induced errors. Finally, we fine-tune the LLM on synthetic noisy examples: we generate imperfect Pinyin-text pairs from intermediate VALLR-Pin checkpoints using the training data, creating instruction-response pairs for error correction. This endows the LLM with awareness of our model's specific error patterns. In summary, VALLR-Pin synergizes visual features with phonetic and linguistic context to improve Mandarin lip-reading performance.

Enhancing Fully Formatted End-to-End Speech Recognition with Knowledge Distillation via Multi-Codebook Vector Quantization

Conventional automatic speech recognition (ASR) models typically produce outputs as normalized texts lacking punctuation and capitalization, necessitating post-processing models to enhance readability. This approach, however, introduces additional complexity and latency due to the cascaded system design. In response to this challenge, there is a growing trend to develop end-to-end (E2E) ASR models capable of directly predicting punctuation and capitalization, though this area remains underexplored. In this paper, we propose an enhanced fully formatted E2E ASR model that leverages knowledge distillation (KD) through multi-codebook vector quantization (MVQ). Experimental results demonstrate that our model significantly outperforms previous works in word error rate (WER) both with and without punctuation and capitalization, and in punctuation error rate (PER). Evaluations on the LibriSpeech-PC test-clean and test-other subsets show that our model achieves state-of-the-art results.

From Speech to Subtitles: Evaluating ASR Models in Subtitling Italian Television Programs

Subtitles are essential for video accessibility and audience engagement. Modern Automatic Speech Recognition (ASR) systems, built upon Encoder-Decoder neural network architectures and trained on massive amounts of data, have progressively reduced transcription errors on standard benchmark datasets. However, their performance in real-world production environments, particularly for non-English content like long-form Italian videos, remains largely unexplored. This paper presents a case study on developing a professional subtitling system for an Italian media company. To inform our system design, we evaluated four state-of-the-art ASR models (Whisper Large v2, AssemblyAI Universal, Parakeet TDT v3 0.6b, and WhisperX) on a 50-hour dataset of Italian television programs. The study highlights their strengths and limitations, benchmarking their performance against the work of professional human subtitlers. The findings indicate that, while current models cannot meet the media industry's accuracy needs for full autonomy, they can serve as highly effective tools for enhancing human productivity. We conclude that a human-in-the-loop (HITL) approach is crucial and present the production-grade, cloud-based infrastructure we designed to support this workflow.