Learnable Pulse Accumulation for On-Device Speech Recognition: How Much Attention Do You Need?

Self-attention scales quadratically with sequence length, limiting transformer-based speech models on edge devices. We introduce the Learnable Pulse Accumulator (LPA), an O(n) replacement that substitutes key-query dot products with learned gating functions: content-dependent rectangular pulses, periodic windows, and position-dependent basis functions. An MSE diagnostic sweep determines per-layer replacement difficulty and ordering. Replacing 8 of 12 wav2vec2-base layers yields 10.61% word error rate (WER) on LibriSpeech test-clean, +7.24 percentage points (pp) over the 3.37% baseline, with 3.27x speedup at 120s audio on Apple M4 Pro via an optimized MLX inference path. Cross-domain validation on SepFormer speech enhancement shows all 16 intra-chunk attention layers can be replaced without collapse, suggesting the depth wall arises from linguistic computation rather than an LPA limitation. LPA's near-binary gates at inference enable dense GPU computation with no CPU-GPU synchronization, and all operations map to mobile neural accelerators.

QV May Be Enough: Toward the Essence of Attention in LLMs

Starting from first principles and a linguistic perspective centered on part-of-speech (POS) and syntactic analysis, this paper explores and derives the underlying essence of the Query-Key-Value (QKV) mechanism within the Transformer architecture. Based on this theoretical foundation, we provide a unified explanatory framework for the efficacy of contemporary architectures, including MQA, GQA, and MLA, while identifying their inherent trade-offs and potential optimization trajectories. We introduce the QV paradigm and provide empirical evidence for its validity. Building upon this, we propose the QV-Ka optimization scheme, which is further substantiated through experimental validation. The interpretable theoretical analysis of the QKV mechanism presented in this work establishes a robust foundation for the future evolution of large language model architectures.

Multi-View Based Audio Visual Target Speaker Extraction

Audio-Visual Target Speaker Extraction (AVTSE) aims to separate a target speaker's voice from a mixed audio signal using the corresponding visual cues. While most existing AVTSE methods rely exclusively on frontal-view videos, this limitation restricts their robustness in real-world scenarios where non-frontal views are prevalent. Such visual perspectives often contain complementary articulatory information that could enhance speech extraction. In this work, we propose Multi-View Tensor Fusion (MVTF), a novel framework that transforms multi-view learning into single-view performance gains. During the training stage, we leverage synchronized multi-perspective lip videos to learn cross-view correlations through MVTF, where pairwise outer products explicitly model multiplicative interactions between different views of input lip embeddings. At the inference stage, the system supports both single-view and multi-view inputs. Experimental results show that in the single-view inputs, our framework leverages multi-view knowledge to achieve significant performance gains, while in the multi-view mode, it further improves overall performance and enhances the robustness. Our demo, code and data are available at https://anonymous.4open.science/w/MVTF-Gridnet-209C/

SimulU: Training-free Policy for Long-form Simultaneous Speech-to-Speech Translation

Simultaneous speech-to-speech translation (SimulS2S) is essential for real-time multilingual communication, with increasing integration into meeting and streaming platforms. Despite this, SimulS2S remains underexplored in research, where current solutions often rely on resource-intensive training procedures and operate on short-form, pre-segmented utterances, failing to generalize to continuous speech. To bridge this gap, we propose SimulU, the first training-free policy for long-form SimulS2S. SimulU adopts history management and speech output selection strategies that exploit cross-attention in pre-trained end-to-end models to regulate both input history and output generation. Evaluations on MuST-C across 8 languages show that SimulU achieves a better or comparable quality-latency trade-off against strong cascaded models. By eliminating the need for ad-hoc training, SimulU offers a promising path to end-to-end SimulS2S in realistic, long-form scenarios.

Beyond Deep Learning: Speech Segmentation and Phone Classification with Neural Assemblies

Deep learning dominates speech processing but relies on massive datasets, global backpropagation-guided weight updates, and produces entangled representations. Assembly Calculus (AC), which models sparse neuronal assemblies via Hebbian plasticity and winner-take-all competition, offers a biologically grounded alternative, yet prior work focused on discrete symbolic inputs. We introduce an AC-based speech processing framework that operates directly on continuous speech by combining three key contributions:(i) neural encoding that converts speech into assembly-compatible spike patterns using probabilistic mel binarisation and population-coded MFCCs; (ii) a multi-area architecture organising assemblies across hierarchical timescales and classes; and (iii) cross-area update schemes for downstream tasks. Applied to two core tasks of boundary detection and segment classification, our framework detects phone (F1=0.69) and word (F1=0.61) boundaries without any weight training, and achieves 47.5% and 45.1% accuracy on phone and command recognition. These results show that AC-based dynamical systems are a viable alternative to deep learning for speech processing.

Duration Aware Scheduling for ASR Serving Under Workload Drift

Scheduling policies in large-scale Automatic Speech Recognition (ASR) serving pipelines play a key role in determining end-to-end (E2E) latency. Yet, widely used serving engines rely on first-come-first-served (FCFS) scheduling, which ignores variability in request duration and leads to head-of-line blocking under workload drift. We show that audio duration is an accurate proxy for job processing time in ASR models such as Whisper, and use this insight to enable duration-aware scheduling. We integrate two classical algorithms, Shortest Job First (SJF) and Highest Response Ratio Next (HRRN), into vLLM and evaluate them under realistic and drifted workloads. On LibriSpeech test-clean, compared to baseline, SJF reduces median E2E latency by up to $73\%$ at high load, but increases $90$th-percentile tail latency by up to $97\%$ due to starvation of long requests. HRRN addresses this trade-off: it reduces median E2E latency by up to $28\%$ while bounding tail-latency degradation to at most $24\%$. These gains persist under workload drift, with no throughput penalty and $<0.1$\,ms scheduling overhead per request.

Synthetic Data Domain Adaptation for ASR via LLM-based Text and Phonetic Respelling Augmentation

End-to-end automatic speech recognition often degrades on domain-specific data due to scarce in-domain resources. We propose a synthetic-data-based domain adaptation framework with two contributions: (1) a large language model (LLM)-based text augmentation pipeline with a filtering strategy that balances lexical diversity, perplexity, and domain-term coverage, and (2) phonetic respelling augmentation (PRA), a novel method that introduces pronunciation variability through LLM-generated orthographic pseudo-spellings. Unlike conventional acoustic-level methods such as SpecAugment, PRA provides phonetic diversity before speech synthesis, enabling synthetic speech to better approximate real-world variability. Experimental results across four domain-specific datasets demonstrate consistent reductions in word error rate, confirming that combining domain-specific lexical coverage with realistic pronunciation variation significantly improves ASR robustness.

When Fine-Tuning Fails and when it Generalises: Role of Data Diversity and Mixed Training in LLM-based TTS

Large language models are increasingly adopted as semantic backbones for neural text-to-speech systems. However, frozen LLM representations are insufficient for modeling speaker specific acoustic and perceptual characteristics. Our experiments involving fine tuning of the Language Model backbone of TTS show promise in improving the voice consistency and Signal to Noise ratio SNR in voice cloning task. Across multiple speakers LoRA finetuning consistently outperforms the non-finetuned base Qwen-0.5B model across three complementary dimensions of speech quality. First, perceptual quality improves significantly with DNS-MOS gains of up to 0.42 points for speakers whose training data exhibits sufficient acoustic variability. Second, speaker fidelity improves for all evaluated speakers with consistent increases in voice similarity indicating that LoRA effectively adapts speaker identity representations without degrading linguistic modeling. Third, signal level quality improves in most cases with signal to noise ratio increasing by as much as 34 percent. Crucially these improvements are strongly governed by the characteristics of the training data. Speakers with high variability in acoustic energy and perceptual quality achieve simultaneous gains in DNS-MOS voice similarity and SNR. Overall this work establishes that LoRA finetuning is not merely a parameter efficient optimization technique but an effective mechanism for better speaker level adaptation in compact LLM-based TTS systems. When supported by sufficiently diverse training data LoRA adapted Qwen-0.5B consistently surpasses its frozen base model in perceptual quality speaker similarity with low latency using GGUF model hosted in quantized form.

Speaker Verification with Speech-Aware LLMs: Evaluation and Augmentation

Speech-aware large language models (LLMs) can accept speech inputs, yet their training objectives largely emphasize linguistic content or specific fields such as emotions or the speaker's gender, leaving it unclear whether they encode speaker identity. First, we propose a model-agnostic scoring protocol that produces continuous verification scores for both API-only and open-weight models, using confidence scores or log-likelihood ratios from the Yes/No token probabilities. Using this protocol, we benchmark recent speech-aware LLMs and observe weak speaker discrimination (EERs above 20% on VoxCeleb1). Second, we introduce a lightweight augmentation that equips an LLM with ASV capability by injecting frozen ECAPA-TDNN speaker embeddings through a learned projection and training only LoRA adapters. On TinyLLaMA-1.1B, the resulting ECAPA-LLM achieves 1.03% EER on VoxCeleb1-E, approaching a dedicated speaker verification system while preserving a natural-language interface.

MoXaRt: Audio-Visual Object-Guided Sound Interaction for XR

In Extended Reality (XR), complex acoustic environments often overwhelm users, compromising both scene awareness and social engagement due to entangled sound sources. We introduce MoXaRt, a real-time XR system that uses audio-visual cues to separate these sources and enable fine-grained sound interaction. MoXaRt's core is a cascaded architecture that performs coarse, audio-only separation in parallel with visual detection of sources (e.g., faces, instruments). These visual anchors then guide refinement networks to isolate individual sources, separating complex mixes of up to 5 concurrent sources (e.g., 2 voices + 3 instruments) with ~2 second processing latency. We validate MoXaRt through a technical evaluation on a new dataset of 30 one-minute recordings featuring concurrent speech and music, and a 22-participant user study. Empirical results indicate that our system significantly enhances speech intelligibility, yielding a 36.2% (p < 0.01) increase in listening comprehension within adversarial acoustic environments while substantially reducing cognitive load (p < 0.001), thereby paving the way for more perceptive and socially adept XR experiences.