Navigating the Reality Gap: Privacy-Preserving Adaptation of ASR for Challenging Low-Resource Domains

Automatic Speech Recognition (ASR) holds immense potential to assist in clinical documentation and patient report generation, particularly in resource-constrained regions. However, deployment is currently hindered by a technical deadlock: a severe "Reality Gap" between laboratory performance and noisy, real-world clinical audio, coupled with strict privacy and resource constraints. We quantify this gap, showing that a robust multilingual model (IndicWav2Vec) degrades to a 40.94% WER on rural clinical data from India, rendering it unusable. To address this, we explore a zero-data-exfiltration framework enabling localized, continual adaptation via Low-Rank Adaptation (LoRA). We conduct a rigorous investigative study of continual learning strategies, characterizing the trade-offs between data-driven and parameter-driven stability. Our results demonstrate that multi-domain Experience Replay (ER) yields the primary performance gains, achieving a 17.1% relative improvement in target WER and reducing catastrophic forgetting by 55% compared to naive adaptation. Furthermore, we observed that standard Elastic Weight Consolidation (EWC) faced numerical stability challenges when applied to LoRA in noisy environments. Our experiments show that a stabilized, linearized formulation effectively controls gradient magnitudes and enables stable convergence. Finally, we verify via a domain-specific spot check that acoustic adaptation is a fundamental prerequisite for usability which cannot be bypassed by language models alone.

TICL+: A Case Study On Speech In-Context Learning for Children's Speech Recognition

Children's speech recognition remains challenging due to substantial acoustic and linguistic variability, limited labeled data, and significant differences from adult speech. Speech foundation models can address these challenges through Speech In-Context Learning (SICL), allowing adaptation to new domains without fine-tuning. However, the effectiveness of SICL depends on how in-context examples are selected. We extend an existing retrieval-based method, Text-Embedding KNN for SICL (TICL), introducing an acoustic reranking step to create TICL+. This extension prioritizes examples that are both semantically and acoustically aligned with the test input. Experiments on four children's speech corpora show that TICL+ achieves up to a 53.3% relative word error rate reduction over zero-shot performance and 37.6% over baseline TICL, highlighting the value of combining semantic and acoustic information for robust, scalable ASR in children's speech.

Phoneme-based speech recognition driven by large language models and sampling marginalization

Recently, the Large Language Model-based Phoneme-to-Grapheme (LLM-P2G) method has shown excellent performance in speech recognition tasks and has become a feasible direction to replace the traditional WFST decoding method. This framework takes into account both recognition accuracy and system scalability through two-stage modeling of phoneme prediction and text generation. However, the existing LLM-P2G adopts the Top-K Marginalized (TKM) training strategy, and its candidate phoneme sequences rely on beam search generation, which has problems such as insufficient path diversity, low training efficiency, and high resource overhead. To this end, this paper proposes a sampling marginalized training strategy (Sampling-K Marginalized, SKM), which replaces beam search with random sampling to generate candidate paths, improving marginalized modeling and training efficiency. Experiments were conducted on Polish and German datasets, and the results showed that SKM further improved the model learning convergence speed and recognition performance while maintaining the complexity of the model. Comparative experiments with a speech recognition method that uses a projector combined with a large language model (SpeechLLM) also show that the SKM-driven LLM-P2G has more advantages in recognition accuracy and structural simplicity. The study verified the practical value and application potential of this method in cross-language speech recognition systems.

Explainable Transformer-CNN Fusion for Noise-Robust Speech Emotion Recognition

Speech Emotion Recognition (SER) systems often degrade in performance when exposed to the unpredictable acoustic interference found in real-world environments. Additionally, the opacity of deep learning models hinders their adoption in trust-sensitive applications. To bridge this gap, we propose a Hybrid Transformer-CNN framework that unifies the contextual modeling of Wav2Vec 2.0 with the spectral stability of 1D-Convolutional Neural Networks. Our dual-stream architecture processes raw waveforms to capture long-range temporal dependencies while simultaneously extracting noise-resistant spectral features (MFCC, ZCR, RMSE) via a custom Attentive Temporal Pooling mechanism. We conducted extensive validation across four diverse benchmark datasets: RAVDESS, TESS, SAVEE, and CREMA-D. To rigorously test robustness, we subjected the model to non-stationary acoustic interference using real-world noise profiles from the SAS-KIIT dataset. The proposed framework demonstrates superior generalization and state-of-the-art accuracy across all datasets, significantly outperforming single-branch baselines under realistic environmental interference. Furthermore, we address the ``black-box" problem by integrating SHAP and Score-CAM into the evaluation pipeline. These tools provide granular visual explanations, revealing how the model strategically shifts attention between temporal and spectral cues to maintain reliability in the presence of complex environmental noise.

Zero-Shot Recognition of Dysarthric Speech Using Commercial Automatic Speech Recognition and Multimodal Large Language Models

Voice-based human-machine interaction is a primary modality for accessing intelligent systems, yet individuals with dysarthria face systematic exclusion due to recognition performance gaps. Whilst automatic speech recognition (ASR) achieves word error rates (WER) below 5% on typical speech, performance degrades dramatically for dysarthric speakers. Multimodal large language models (MLLMs) offer potential for leveraging contextual reasoning to compensate for acoustic degradation, yet their zero-shot capabilities remain uncharacterised. This study evaluates eight commercial speech-to-text services on the TORGO dysarthric speech corpus: four conventional ASR systems (AssemblyAI, Whisper large-v3, Deepgram Nova-3, Nova-3 Medical) and four MLLM-based systems (GPT-4o, GPT-4o Mini, Gemini 2.5 Pro, Gemini 2.5 Flash). Evaluation encompasses lexical accuracy, semantic preservation, and cost-latency trade-offs. Results demonstrate severity-dependent degradation: mild dysarthria achieves 3-5% WER approaching typical-speech benchmarks, whilst severe dysarthria exceeds 49% WER across all systems. A verbatim-transcription prompt yields architecture-specific effects: GPT-4o achieves 7.36 percentage point WER reduction with consistent improvement across all tested speakers, whilst Gemini variants exhibit degradation. Semantic metrics indicate that communicative intent remains partially recoverable despite elevated lexical error rates. These findings establish empirical baselines enabling evidence-based technology selection for assistive voice interface deployment.

Incorporating Error Level Noise Embedding for Improving LLM-Assisted Robustness in Persian Speech Recognition

Automatic Speech Recognition (ASR) systems suffer significant performance degradation in noisy environments, a challenge that is especially severe for low-resource languages such as Persian. Even state-of-the-art models such as Whisper struggle to maintain accuracy under varying signal-to-noise ratios (SNRs). This study presents a robust noise-sensitive ASR error correction framework that combines multiple hypotheses and noise-aware modeling. Using noisy Persian speech, we generate 5-best hypotheses from a modified Whisper-large decoder. Error Level Noise (ELN) is introduced as a representation that captures semantic- and token-level disagreement across hypotheses, quantifying the linguistic distortions caused by noise. ELN thus provides a direct measure of noise-induced uncertainty, enabling the LLM to reason about the reliability of each hypothesis during correction. Three models are evaluated: (1) a base LLaMA-2-7B model without fine-tuning, (2) a fine-tuned variant trained on text-only hypotheses, and (3) a noise-conditioned model integrating ELN embeddings at both sentence and word levels. Experimental results demonstrate that the ELN-conditioned model achieves substantial reductions in Word Error Rate (WER). Specifically, on the challenging Mixed Noise test set, the proposed Fine-tuned + ELN (Ours) model reduces the WER from a baseline of 31.10\% (Raw Whisper) to 24.84\%, significantly surpassing the Fine-tuned (No ELN) text-only baseline of 30.79\%, whereas the original LLaMA-2-7B model increased the WER to 64.58\%, demonstrating that it is unable to correct Persian errors on its own. This confirms the effectiveness of combining multiple hypotheses with noise-aware embeddings for robust Persian ASR in noisy real-world scenarios.

Peeking Into The Future For Contextual Biasing

While end-to-end (E2E) automatic speech recognition (ASR) models excel at general transcription, they struggle to recognize rare or unseen named entities (e.g., contact names, locations), which are critical for downstream applications like virtual assistants. In this paper, we propose a contextual biasing method for attention based encoder decoder (AED) models using a list of candidate named entities. Instead of predicting only the next token, we simultaneously predict multiple future tokens, enabling the model to "peek into the future" and score potential candidate entities in the entity list. Moreover, our approach leverages the multi-token prediction logits directly without requiring additional entity encoders or cross-attention layers, significantly reducing architectural complexity. Experiments on Librispeech demonstrate that our approach achieves up to 50.34% relative improvement in named entity word error rate compared to the baseline AED model.

When De-noising Hurts: A Systematic Study of Speech Enhancement Effects on Modern Medical ASR Systems

Speech enhancement methods are commonly believed to improve the performance of automatic speech recognition (ASR) in noisy environments. However, the effectiveness of these techniques cannot be taken for granted in the case of modern large-scale ASR models trained on diverse, noisy data. We present a systematic evaluation of MetricGAN-plus-voicebank denoising on four state-of-the-art ASR systems: OpenAI Whisper, NVIDIA Parakeet, Google Gemini Flash 2.0, Parrotlet-a using 500 medical speech recordings under nine noise conditions. ASR performance is measured using semantic WER (semWER), a normalized word error rate (WER) metric accounting for domain-specific normalizations. Our results reveal a counterintuitive finding: speech enhancement preprocessing degrades ASR performance across all noise conditions and models. Original noisy audio achieves lower semWER than enhanced audio in all 40 tested configurations (4 models x 10 conditions), with degradations ranging from 1.1% to 46.6% absolute semWER increase. These findings suggest that modern ASR models possess sufficient internal noise robustness and that traditional speech enhancement may remove acoustic features critical for ASR. For practitioners deploying medical scribe systems in noisy clinical environments, our results indicate that preprocessing audio with noise reduction techniques might not just be computationally wasteful but also be potentially harmful to the transcription accuracy.

Scalable Frameworks for Real-World Audio-Visual Speech Recognition

The practical deployment of Audio-Visual Speech Recognition (AVSR) systems is fundamentally challenged by significant performance degradation in real-world environments, characterized by unpredictable acoustic noise and visual interference. This dissertation posits that a systematic, hierarchical approach is essential to overcome these challenges, achieving the robust scalability at the representation, architecture, and system levels. At the representation level, we investigate methods for building a unified model that learns audio-visual features inherently robust to diverse real-world corruptions, thereby enabling generalization to new environments without specialized modules. To address architectural scalability, we explore how to efficiently expand model capacity while ensuring the adaptive and reliable use of multimodal inputs, developing a framework that intelligently allocates computational resources based on the input characteristics. Finally, at the system level, we present methods to expand the system's functionality through modular integration with large-scale foundation models, leveraging their powerful cognitive and generative capabilities to maximize final recognition accuracy. By systematically providing solutions at each of these three levels, this dissertation aims to build a next-generation, robust, and scalable AVSR system with high reliability in real-world applications.

Adaptive Edge-Cloud Inference for Speech-to-Action Systems Using ASR and Large Language Models

Voice-based interaction has emerged as a natural and intuitive modality for controlling IoT devices. However, speech-driven edge devices face a fundamental trade-off between cloud-based solutions, which offer stronger language understanding capabilities at the cost of latency, connectivity dependence, and privacy concerns, and edge-based solutions, which provide low latency and improved privacy but are limited by computational constraints. This paper presents ASTA, an adaptive speech-to-action solution that dynamically routes voice commands between edge and cloud inference to balance performance and system resource utilization. ASTA integrates on-device automatic speech recognition and lightweight offline language-model inference with cloud-based LLM processing, guided by real-time system metrics such as CPU workload, device temperature, and network latency. A metric-aware routing mechanism selects the inference path at runtime, while a rule-based command validation and repair component ensures successful end-to-end command execution. We implemented our solution on an NVIDIA Jetson-based edge platform and evaluated it using a diverse dataset of 80 spoken commands. Experimental results show that ASTA successfully routes all input commands for execution, achieving a balanced distribution between online and offline inference. The system attains an ASR accuracy of 62.5% and generates executable commands without repair for only 47.5% of inputs, highlighting the importance of the repair mechanism in improving robustness. These results suggest that adaptive edge-cloud orchestration is a viable approach for resilient and resource-aware voice-controlled IoT systems.