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.

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.

Reproducing and Dissecting Denoising Language Models for Speech Recognition

Denoising language models (DLMs) have been proposed as a powerful alternative to traditional language models (LMs) for automatic speech recognition (ASR), motivated by their ability to use bidirectional context and adapt to a specific ASR model's error patterns. However, the complexity of the DLM training pipeline has hindered wider investigation. This paper presents the first independent, large-scale empirical study of DLMs. We build and release a complete, reproducible pipeline to systematically investigate the impact of key design choices. We evaluate dozens of configurations across multiple axes, including various data augmentation techniques (e.g., SpecAugment, dropout, mixup), different text-to-speech systems, and multiple decoding strategies. Our comparative analysis in a common subword vocabulary setting demonstrates that DLMs outperform traditional LMs, but only after a distinct compute tipping point. While LMs are more efficient at lower budgets, DLMs scale better with longer training, mirroring behaviors observed in diffusion language models. However, we observe smaller improvements than those reported in prior character-based work, which indicates that the DLM's performance is conditional on factors such as the vocabulary. Our analysis reveals that a key factor for improving performance is to condition the DLM on richer information from the ASR's hypothesis space, rather than just a single best guess. To this end, we introduce DLM-sum, a novel method for decoding from multiple ASR hypotheses, which consistently outperforms the previously proposed DSR decoding method. We believe our findings and public pipeline provide a crucial foundation for the community to better understand, improve, and build upon this promising class of models. The code is publicly available at https://github.com/rwth-i6/2025-denoising-lm/.

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.

A stylometric analysis of speaker attribution from speech transcripts

Forensic scientists often need to identify an unknown speaker or writer in cases such as ransom calls, covert recordings, alleged suicide notes, or anonymous online communications, among many others. Speaker recognition in the speech domain usually examines phonetic or acoustic properties of a voice, and these methods can be accurate and robust under certain conditions. However, if a speaker disguises their voice or employs text-to-speech software, vocal properties may no longer be reliable, leaving only their linguistic content available for analysis. Authorship attribution methods traditionally use syntactic, semantic, and related linguistic information to identify writers of written text (authorship attribution). In this paper, we apply a content-based authorship approach to speech that has been transcribed into text, using what a speaker says to attribute speech to individuals (speaker attribution). We introduce a stylometric method, StyloSpeaker, which incorporates character, word, token, sentence, and style features from the stylometric literature on authorship, to assess whether two transcripts were produced by the same speaker. We evaluate this method on two types of transcript formatting: one approximating prescriptive written text with capitalization and punctuation and another normalized style that removes these conventions. The transcripts' conversation topics are also controlled to varying degrees. We find generally higher attribution performance on normalized transcripts, except under the strongest topic control condition, in which overall performance is highest. Finally, we compare this more explainable stylometric model to black-box neural approaches on the same data and investigate which stylistic features most effectively distinguish speakers.

EEG-to-Voice Decoding of Spoken and Imagined speech Using Non-Invasive EEG

Restoring speech communication from neural signals is a central goal of brain-computer interface research, yet EEG-based speech reconstruction remains challenging due to limited spatial resolution, susceptibility to noise, and the absence of temporally aligned acoustic targets in imagined speech. In this study, we propose an EEG-to-Voice paradigm that directly reconstructs speech from non-invasive EEG signals without dynamic time warping (DTW) or explicit temporal alignment. The proposed pipeline generates mel-spectrograms from EEG in an open-loop manner using a subject-specific generator, followed by pretrained vocoder and automatic speech recognition (ASR) modules to synthesize speech waveforms and decode text. Separate generators were trained for spoken speech and imagined speech, and transfer learning-based domain adaptation was applied by pretraining on spoken speech and adapting to imagined speech. A minimal language model-based correction module was optionally applied to correct limited ASR errors while preserving semantic structure. The framework was evaluated under 2 s and 4 s speech conditions using acoustic-level metrics (PCC, RMSE, MCD) and linguistic-level metrics (CER, WER). Stable acoustic reconstruction and comparable linguistic accuracy were observed for both spoken speech and imagined speech. While acoustic similarity decreased for longer utterances, text-level decoding performance was largely preserved, and word-position analysis revealed a mild increase in decoding errors toward later parts of sentences. The language model-based correction consistently reduced CER and WER without introducing semantic distortion. These results demonstrate the feasibility of direct, open-loop EEG-to-Voice reconstruction for spoken speech and imagined speech without explicit temporal alignment.

All-in-One ASR: Unifying Encoder-Decoder Models of CTC, Attention, and Transducer in Dual-Mode ASR

This paper proposes a unified framework, All-in-One ASR, that allows a single model to support multiple automatic speech recognition (ASR) paradigms, including connectionist temporal classification (CTC), attention-based encoder-decoder (AED), and Transducer, in both offline and streaming modes. While each ASR architecture offers distinct advantages and trade-offs depending on the application, maintaining separate models for each scenario incurs substantial development and deployment costs. To address this issue, we introduce a multi-mode joiner that enables seamless integration of various ASR modes within a single unified model. Experiments show that All-in-One ASR significantly reduces the total model footprint while matching or even surpassing the recognition performance of individually optimized ASR models. Furthermore, joint decoding leverages the complementary strengths of different ASR modes, yielding additional improvements in recognition accuracy.

TRIDENT: A Redundant Architecture for Caribbean-Accented Emergency Speech Triage

Emergency speech recognition systems exhibit systematic performance degradation on non-standard English varieties, creating a critical gap in services for Caribbean populations. We present TRIDENT (Transcription and Routing Intelligence for Dispatcher-Empowered National Triage), a three-layer dispatcher-support architecture designed to structure emergency call inputs for human application of established triage protocols (the ESI for routine operations and START for mass casualty events), even when automatic speech recognition fails. The system combines Caribbean-accent-tuned ASR, local entity extraction via large language models, and bio-acoustic distress detection to provide dispatchers with three complementary signals: transcription confidence, structured clinical entities, and vocal stress indicators. Our key insight is that low ASR confidence, rather than representing system failure, serves as a valuable queue prioritization signal -- particularly when combined with elevated vocal distress markers indicating a caller in crisis whose speech may have shifted toward basilectal registers. A complementary insight drives the entity extraction layer: trained responders and composed bystanders may report life-threatening emergencies without elevated vocal stress, requiring semantic analysis to capture clinical indicators that paralinguistic features miss. We describe the architectural design, theoretical grounding in psycholinguistic research on stress-induced code-switching, and deployment considerations for offline operation during disaster scenarios. This work establishes a framework for accent-resilient emergency AI that ensures Caribbean voices receive equitable access to established national triage protocols. Empirical validation on Caribbean emergency calls remains future work.

GeoSense-AI: Fast Location Inference from Crisis Microblogs

This paper presents an applied AI pipeline for realtime geolocation from noisy microblog streams, unifying statistical hashtag segmentation, part-of-speech-driven proper-noun detection, dependency parsing around disaster lexicons, lightweight named-entity recognition, and gazetteer-grounded disambiguation to infer locations directly from text rather than sparse geotags. The approach operationalizes information extraction under streaming constraints, emphasizing low-latency NLP components and efficient validation against geographic knowledge bases to support situational awareness during emergencies. In head to head comparisons with widely used NER toolkits, the system attains strong F1 while being engineered for orders-of-magnitude faster throughput, enabling deployment in live crisis informatics settings. A production map interface demonstrates end-to-end AI functionality ingest, inference, and visualization--surfacing locational signals at scale for floods, outbreaks, and other fastmoving events. By prioritizing robustness to informal text and streaming efficiency, GeoSense-AI illustrates how domain-tuned NLP and knowledge grounding can elevate emergency response beyond conventional geo-tag reliance.

Robust Speech Activity Detection in the Presence of Singing Voice

Speech Activity Detection (SAD) systems often misclassify singing as speech, leading to degraded performance in applications such as dialogue enhancement and automatic speech recognition. We introduce Singing-Robust Speech Activity Detection ( SR-SAD ), a neural network designed to robustly detect speech in the presence of singing. Our key contributions are: i) a training strategy using controlled ratios of speech and singing samples to improve discrimination, ii) a computationally efficient model that maintains robust performance while reducing inference runtime, and iii) a new evaluation metric tailored to assess SAD robustness in mixed speech-singing scenarios. Experiments on a challenging dataset spanning multiple musical genres show that SR-SAD maintains high speech detection accuracy (AUC = 0.919) while rejecting singing. By explicitly learning to distinguish between speech and singing, SR-SAD enables more reliable SAD in mixed speech-singing scenarios.