Abstract

Text-to-speech systems are continuously improving, making the evaluation of perceptual quality increasingly challenging. Existing MOS prediction methods reduce speech quality to a single scalar value, without capturing how different listeners perceive audio across dimensions like acoustic fidelity, prosodic naturalness, phonetic intelligibility, and semantic coherence. We present MDMOS, a multi-dimensional framework where audio is processed through four complementary perceptual expert heads, each combining layer-grouped WavLM-Large representations with hand-crafted signal-processing features. An ordinal KL loss provides each head with independent supervision, preventing head collapse. A quality-adaptive fusion module dynamically weights the expert scores, while orthogonality regularisation and variance-weighted training account for rater uncertainty. Experiments on SOMOS and BVCC demonstrate strong MOS prediction accuracy, achieving system-level SRCC of 0.925 and competitive performance with 0.944 respectively.

Abstract

Text-to-Speech (TTS) technology has achieved remarkable progress with the advent of deep learning, low-resource indigenous languages still remain at the margins of this advancement. In the Indian context, indigenous languages such as Bhili, Gondi, Santhali, and Mundari, spoken by millions of people, are critically underrepresented. This paper addresses this gap by developing neural TTS systems for these languages through the creation of a comprehensive speech-text paired corpus combining translated text from resource-rich languages and original compositions in indigenous languages. We trained and fine-tuned Glow-TTS models using baseline and transfer learning approaches. Subjective and objective evaluation metrics show that Bhili achieved the highest Mean Opinion Score (MOS) when fine-tuned from a Hindi-pretrained model. Gondi, Santhali, and partly Mundari performed better when trained from scratch, underscoring the need for language-specific modeling in structurally diverse languages.

Abstract

This paper presents a novel framework for fine-grained classification of emphasis in spontaneous speech by moving beyond traditional binary stress detection, which rely on handcrafted features like MFCCs, pitch and formant frequencies. By utilizing a lecture mode multi-speaker speech corpus annotated with two stress levels: `High' and `Low', we propose a methodology that utilises a fusion of acoustic embeddings along with contextual linguistic features from pretrained Self-Supervised models. This feature set is processed by a Temporal CNN-BiLSTM network to learn the intricate temporal patterns of stress. The results demonstrate that the proposed approach outperforms the traditional baselines with an accuracy of 98% on the Stressed vs Unstressed task and 96% on the Low vs High-stress task. The framework achieved an accuracy of 95% on the multi-class (Unstressed vs Low vs High) task, validating its ability to perform real-world nuanced detection of emphasis at varied levels and speakers.

Abstract

This review paper discusses the evolution of Integrated Modular Avionics (IMA) architecture for aerospace applications and the adoption of IMA in missile applications. It provides an overview of the traditional Federated Architecture and also explains the concept of Integrated Modular Avionics (IMA) architecture. The paper also discusses about the design, implementation, methodology, and compares the pros and cons of both architectures. It also highlights the characteristics of IMA, the design challenges, and how these challenges are overcome by providing solutions. The methods adopted in building IMA systems and the improvements made are clearly explained. This review paper presents the results of a pilot implementation and provides recommendations for future adoption of this technology in missile applications. The paper also discusses about the lessons learned from various commercial IMA systems (worldwide), these learnings can be adopted for successful IMA system development for military applications.

Abstract

Speech enhancement models optimised for percep-
tual quality metrics such as PESQ do not necessarily improve
automatic speech recognition (ASR) performance, as aggres-
sive noise suppression can distort the phonemic cues that
acoustic models rely on. We propose a time-domain speech
frontend based on the Mamba selective state space model,
trained exclusively with ASR-oriented loss functions to pre-
serve phonetically discriminative structure rather than max-
imise perceptual quality. The model follows a U-Net encoder-
decoder architecture with 4× temporal downsampling and six
stacked bidirectional Mamba blocks operating on the raw wave-
form, trained with a multi-component loss combining multi-
resolution spectral supervision, log-mel feature matching, speech
correlation preservation, and ideal-ratio-mask spectral objec-
tives. On VoiceBank-DEMAND, the proposed model achieves
10.26% WER with wav2vec2-base-960h and 7.68% with
whisper-base, achieving statistically equivalent WER to SE-
Mamba (10.30% / 7.73%) while producing a lower PESQ
of 3.10 versus 3.69. Phoneme-class analysis confirms that the
proposed model matches or outperforms SEMamba on five
of seven articulatory classes, demonstrating that ASR-oriented
training objectives achieve equivalent robustness to perceptual
enhancement models without modifying the downstream ASR
system.

Abstract

Developing Automatic Speech Recognition (ASR)
systems for specialized medical domains is challenging in low-
resource settings due to the limited availability of annotated
speech data. Synthetic speech generated using text to speech
(TTS) systems is often used to augment training data, but directly
mixing synthetic and real speech can introduce distribution
mismatch that degrades encoder representations. This work
introduces Real-Governed Representation-Calibrated Training
(RG-RCT), a training strategy that regulates the influence of syn-
thetic speech at the representation level. The proposed framework
combines reliability-based confidence estimation, representation
alignment, and layer-wise representation governance to stabilize
the encoder feature space while preserving acoustic diversity from
multi-speaker synthetic data. Experiments on medical-domain
speech datasets using Wav2Vec2 and Whisper demonstrate that
RG-RCT consistently outperforms conventional strategies such
as direct mixing and confidence-aware training. The proposed
method achieves the lowest Word Error Rate (WER), reaching
17.1% and 16.6% for Telugu and 23.3% and 25.6% for Kannada.
These results indicate that regulating encoder representation
geometry enables more effective utilization of synthetic speech
for low-resource medical ASR

Abstract

Text-to-speech systems are improving fast, but measuring how natural they sound still requires expensive human listening tests; existing automatic methods struggle to generalise well across different datasets, so we present QMOS, a MOS prediction framework that extracts hierarchical speech quality features from a frozen Qwen2-Audio large audio language model and combines them with layer-weighted WavLM SSL representations through a learned cross-attention fusion, capturing both high-level semantic naturalness and low-level acoustic distortions in a unified model, and on two standard benchmarks, SOMOS and BVCC, QMOS achieves competitive system-level SRCC of 0.923 and 0.921 respectively while using no system-ID conditioning or listener embeddings, with cross-domain evaluation yielding a competitive SRCC of 0.702, suggesting the learned representations generalise across acoustic domains.

Abstract

Current systems for stuttering detection are limited by task-agnostic feature compression and flat layer-fusion strategies. Traditional methods rely on Principal Component Analysis (PCA) to reduce self-supervised representations, which often suppresses subtle disfluency patterns in favor of speaker-identity cues. We propose ResGDS-HLF, a framework that replaces static compression with Residual Gumbel Dimension Selection to recover discriminative dimensions directly optimized for stuttering classification. By partitioning the transformer backbone into a functional hierarchy, i.e.~acoustic, phonetic, and lexical, the model preserves specialized information that is otherwise diluted by weighted averaging. Evaluated on the SEP-28K corpus, ResGDS-HLF achieves a Macro-F1 of 0.61, representing an absolute gain of 0.14 over single-layer PCA baselines. Zero-shot transfer to FluencyBank confirms that task-driven Gumbel selection captures robust, generalizable disfluency signatures across diverse acoustic environments.

Abstract

Reservoir Computing (RC) has gained increasing
attention for its efficiency and adaptability in processing time-series and other complex temporal data. While software simulations can easily support large reservoirs by mapping inputs into high-dimensional spaces, hardware implementations remain challenging due to architectural complexity. Most RC systems adopt sequential structures, which simplify hardware but sacrifice the low latency and inherent fault tolerance of parallel architectures. Although innovative memory nonlinearity trade-off mechanisms have been proposed, their integration into hardware has not yet been fully realized. In addition, the widely used hyperbolic tangent activation function is computationally expensive, whereas simplified variants like the Tanh-Linear function lack sufficient nonlinearity, resulting in reduced accuracy. This work presents a fully parallel hardware implementation of the Simple Cycle Reservoir with a novel activation function. A genetic algorithm is integrated and automated to tune the memory nonlinearity trade-off factor, leakage rate, and several other hyperparameters, enabling seamless deployment into hardware. Our unified hardware-software co-optimization framework provides a low- power and high throughput fully parallel reservoir implementation with excellent accuracy. Experiments on standard bench-
mark time-series datasets demonstrate up to a tenfold reduction in prediction error, a 10-100 times improvement in speed, a power-delay product below 0.02 W·μs, and LUT utility to 2% of conventional reservoir implementation with minimal effect on DSP utility.

Abstract

Automatic Speech Recognition (ASR) for low resource Dravidian languages like Telugu and Kannada faces significant challenges in specialized medical domains due to limited annotated data and morphological complexity. This work proposes a novel confidence-aware training framework that integrates real and synthetic speech data through a hybrid confidence mechanism combining static perceptual and acoustic similarity metrics with dynamic model entropy. Unlike direct fine-tuning approaches, the proposed methodology employs both fixed-weight and learnable-weight confidence aggregation strategies to guide sample weighting during training, enabling effective utilization of heterogeneous data sources. The framework is evaluated on Telugu and Kannada medical datasets containing both real recordings and TTS-generated synthetic speech. A 5gram KenLM language model is applied for post-decoding correction. Results show that the hybrid confidence-aware approach with learnable weights substantially reduces recognition errors: Telugu Word Error Rate (WER) decreases from 24.3 % to 15.8 % (8.5 % absolute improvement), while Kannada WER drops from 31.7% to 25.4% (6.3% absolute improvement), both significantly outperforming standard fine-tuning baselines. These findings confirm that combining adaptive confidence-aware training with statistical language modeling delivers superior performance for domain-specific ASR in morphologically complex Dravidian languages.