Speech enhancement models optimised for perceptual quality metrics such as PESQ do not necessarily improve automatic speech recognition (ASR) performance, as aggressive 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 preserve phonetically discriminative structure rather than maximise perceptual quality. The model follows a U-Net encoderdecoder architecture with 4 × temporal downsampling and six stacked bidirectional Mamba blocks operating on the raw waveform, trained with a multi-component loss combining multiresolution spectral supervision, log-mel feature matching, speech correlation preservation, and ideal-ratio-mask spectral objectives. 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.

Automatic stuttering detection is hindered by two major challenges: extreme class imbalance, where minority stutter types represent less than five percent of the data, and long-range temporal dependencies that span multiple non-adjacent speech segments. We address these issues with HyDRA, a hierarchical hypergraph neural network that factorizes the detection task into binary disfluency recognition followed by subtype classification. HyDRA constructs view-specific hypergraphs from wav2vec2 and HuBERT embeddings, where hyperedges connect multiple acoustically similar segments, enabling modeling of repetitive and prosodic patterns that pairwise graphs cannot represent. Experiments on the SEP-28k dataset show that HyDRA achieves 47.2 macro F1 for subtype classification, compared to 31.4 for flat baselines, with gains exceeding 16 points on minority classes. Cross-dataset evaluation on FluencyBank confirms generalization across domains. These results demonstrate that hierarchical factorization, higher-order hypergraph modeling, and structured multi-view integration provide a principled solution to the imbalance and dependency challenges in automatic stuttering detection.

Automatic Speech Recognition (ASR) for lowresource 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.

Developing Automatic Speech Recognition (ASR) systems for specialized medical domains is challenging in lowresource 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 synthetic 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.

Despite advancements in Speech-to-Speech Translation, maintaining expressiveness between the source and target speech, particularly between English and Indian languages, remains challenging. This study investigates prosodic similarities and variations between English and Malayalam, a language spoken in southern India. A set of 22 prompts from the IViE corpus, covering five categories-simple sentences, WH-questions, questions without morphosyntactic markers, inversion questions, and coordinations-was selected. These prompts, originally spoken by UK speakers, were translated into Malayalam and both language prompts were recorded by bilingual Malayalam speakers while preserving expressiveness. Word-level prominence was manually annotated, and comparisons were made across Indian English, Malayalam, and UK English. The analysis reveals that prominence is retained at key points in Indian English, whereas in Malayalam, it is low due to question specific diacritics and agglutinative nature.

Automatic detection of prominence at the word and syllable-levels is critical for building computer-assisted language learning systems. It has been shown that prosody embeddings learned by the current state-of-the-art (SOTA) text-to-speech (TTS) systems could generate word- and syllable-level prominence in the synthesized speech as natural as in native speech. To understand the effectiveness of prosody embeddings from TTS for prominence detection under nonnative context, a comparative analysis is conducted on the embeddings extracted from native and non-native speech considering the prominence-related embeddings: duration, energy, and pitch from a SOTA TTS named FastSpeech2. These embeddings are extracted under two conditions considering: 1) only text, 2) both speech and text. For the first condition, the embeddings are extracted directly from the TTS inference mode, whereas for the second condition, we propose to extract from the TTS under training mode. Experiments are conducted on native speech corpus: Tatoeba, and non-native speech corpus: ISLE. For experimentation, word-level prominence locations are manually annotated for both corpora. The highest relative improvement on word & syllable-level prominence detection accuracies with the TTS embeddings are found to be 13.7% & 5.9% and 16.2% & 6.9% compared to those with the heuristic-based features and self-supervised Wav2Vec-2.0 representations, respectively.

Traditionally, dysarthric speech intelligibility assessment systems have focused on speech as the primary input, utilizing methods such as extraction of relevant speech features, classification models, alignment of Automatic Speech Recognition (ASR) outputs, and comparisons between speech representations of dysarthric and healthy speakers. However, to achieve an automated intelligibility assessment that closely mirrors the auditory-perceptual evaluations conducted by clinicians, a model that captures both the acoustic characteristics of dysarthric speech and the linguistic structure related to word pronunciation are needed. Inspired by the practices of clinicians, this study introduces a novel text-guided dysarthric speech intelligibility assessment framework that leverages custom keyword spotting (DySIA-CKWS). The model evaluates intelligibility by detecting specific keywords and is extensively tested using UA-Speech database for speaker-wise analysis and across word groups of varying complexity. To ensure robustness, the system's performance is further validated on TORGO database, demonstrating its adaptability in cross-database settings. Statistical analysis demonstrates strong alignment between predicted and subjective intelligibility scores, with a Pearson Correlation Coefficient (PCC) of 0.9588 and a Spearman's Correlation Coefficient (SCC) of 0.9141, achieved using the proposed system on the UA-Speech database. The findings emphasize the importance of word selection and showcase the model's effectiveness in diagnosing dysarthric speech, offering a significant advancement in intelligibility assessment methodologies.

Dysarthric speech poses significant challenges in developing assistive technologies, primarily due to the limited availability of data. Recent advances in neural speech synthesis, especially zero-shot voice cloning, facilitate synthetic speech generation for data augmentation; however, they may introduce biases towards dysarthric speech. In this paper, we investigate the effectiveness of state-of-the-art F5-TTS in cloning dysarthric speech using TORGO dataset, focusing on intelligibility, speaker similarity, and prosody preservation. We also analyze potential biases using fairness metrics like Disparate Impact and Parity Difference to assess disparities across dysarthric severity levels. Results show that F5-TTS exhibits a strong bias toward speech intelligibility over speaker and prosody preservation in dysarthric speech synthesis. Insights from this study can help integrate fairness-aware dysarthric speech synthesis, fostering the advancement of more inclusive speech technologies.

Automatic detection and severity assessment of dysarthria are crucial for delivering targeted therapeutic interventions to patients. While most existing research focuses primarily on speech modality, this study introduces a novel approach that leverages both speech and text modalities. By employing cross-attention mechanism, our method learns the acoustic and linguistic similarities between speech and text representations. This approach assesses specifically the pronunciation deviations across different severity levels, thereby enhancing the accuracy of dysarthric detection and severity assessment. All the experiments have been performed using UA-Speech dysarthric database. Improved accuracies of 99.53% and 93.20% in detection, and 98.12% and 51.97% for severity assessment have been achieved when speaker-dependent and speaker-independent, unseen and seen words settings are used. These findings suggest that by integrating text information, which provides a reference linguistic knowledge, a more robust framework has been developed for dysarthric detection and assessment, thereby potentially leading to more effective diagnoses.

This study primarily aims to enhance an English-to-Hindi Speech-to-Speech Machine Translation (SSMT) system that can improve the delivery of educational content by incorporating stress cues into the translated speech and increasing listener engagement and interest. Frame-level features such as F0, energy, Mel Frequency Cepstral Coefficients and Shifted Delta Coefficients are extracted and fed into StressNet, a Time Delay Neural Network-based stress detection model, which evaluates the presence of stress in individual frames, subsequently synthesizing these frame-level assessments into an aggregated word-level prediction. To enable the incorporation of stress in the generated speech in the target language, variance predictors inspired by FastPitch are added to the YourTTS-based end-to-end Text-to-Speech (TTS) architecture. Comprehensive evaluations were conducted on both the stress detection model and the modified TTS system, which helped identify the most effective configurations for achieving optimal performance. The paper demonstrates the effectiveness of this approach, presenting a refined English-to-Hindi SSMT system that enhances stress transfer ability in the target speech.

This paper presents the submission of IIITH-BUT to the IWSLT 2025 shared task on speech translation for the low-resource Bhojpuri-Hindi language pair. We explored the impact of hyperparameter optimisation and data augmentation techniques on the performance of the SeamlessM4T model fine-tuned for this specific task. We systematically investigated a range of hyperparameters including learning rate schedules, number of update steps, warm-up steps, label smoothing, and batch sizes; and report their effect on translation quality. To address data scarcity, we applied speed perturbation and SpecAugment and studied their effect on translation quality. We also examined the use of cross-lingual signal through joint training with Marathi and Bhojpuri speech data. Our experiments reveal that careful selection of hyperparameters and the application of simple yet effective augmentation techniques significantly improve performance in low-resource settings. We also analysed the translation hypotheses to understand various kinds of errors that impacted the translation quality in terms of BLEU.

The scarcity of high-quality annotated data presents a significant challenge in developing effective end-to-end speech-to-text translation (ST) systems, particularly for low-resource languages. This paper explores the hypothesis that weakly labeled data can be used to build ST models for low-resource language pairs. We constructed speech-to-text translation datasets with the help of bitext mining using state-of-the-art sentence encoders. We mined the multilingual Shrutilipi corpus to build Shrutilipi-anuvaad, a dataset comprising ST data for language pairs Bengali-Hindi, Malayalam-Hindi, Odia-Hindi, and Telugu-Hindi. We created multiple versions of training data with varying degrees of quality and quantity to investigate the effect of quality versus quantity of weakly labeled data on ST model performance. Results demonstrate that ST systems can be built using weakly labeled data, with performance comparable to massive multi-modal multilingual baselines such as SONAR and SeamlessM4T.

Speech disfluencies in spontaneous communication can be categorized as either typical or atypical. Typical disfluencies, such as hesitations and repetitions, are natural occurrences in everyday speech, while atypical disfluencies are indicative of pathological disorders like stuttering. Distinguishing between these categories is crucial for improving voice assistants (VAs) for Persons Who Stutter (PWS), who often face premature cutoffs due to misidentification of speech termination. Accurate classification also aids in detecting stuttering early in children, preventing misdiagnosis as language development disfluency. This research introduces the IIITH-TISA dataset, the first Indian English stammer corpus, capturing atypical disfluencies. Additionally, we extend the IIITH-IED dataset with detailed annotations for typical disfluencies. We propose Perceptually Enhanced Zero-Time Windowed Cepstral Coefficients (PE-ZTWCC) combined with Shifted Delta Cepstra (SDC) as input features to a shallow Time Delay Neural Network (TDNN) classifier, capturing both local and wider temporal contexts. Our method achieves an average F1 score of 85.01% for disfluency classification, outperforming traditional features.

This paper investigates the nuanced distinctions between typical and atypical speech disfluencies, focusing on features captured by self-supervised models. Typical disfluencies are natural, non-pathological irregularities in speech, while atypical disfluencies are linked to speech disorders like stuttering, characterized by more frequent and severe disruptions. Despite progress in automatic disfluency detection, limited research addresses the direct classification of these two types. This study leverages intermediate representations from four pre-trained models Wav2Vec2.0, HuBERT, WavLM, and TERA, to analyze and classify typical and stuttered disfluencies. The experiments utilize two novel Indian English datasets, IIITH-IED-E and IIITH-TISA, enabling a comprehensive analysis of disfluency patterns in a linguistically diverse context. Classification experiments with support vector machines (SVM) and convolutional neural networks (CNN) reveal that features from HuBERT's 5th layer-balancing low-level acoustic and high-level semantic information-achieve a peak F1 score of 0.97. These findings highlight the importance of intermediate layer representations of self-supervised models in distinguishing nuanced speech variations and contribute to robust and interpretable systems for speech disfluency classification.

Automatic detection and classification of stuttering events remain a challenging problem in speech processing, particularly due to the variability of disfluency (In this work, the terms *stuttering* and *disfluency* are used interchangeably) types and the scarcity of large annotated datasets. This paper presents a signal processing-based approach for stuttering classification that avoids the data-intensive requirements of machine learning models. Our method focuses on syllable-level analysis using key acoustic parameters such as energy patterns, pitch contours, spectral stability, and temporal features, implemented through rule-based classification with predefined thresholds. The methodology involves automatic syllable segmentation followed by rule-based classification using thresholds on acoustic features. The dataset comprises annotated read and spontaneous speech from 106 Kannada-speaking adults, including 26 new participants evaluated during clinical trials. Three certified Speech-Language Pathologists (SLPs) conducted perceptual evaluations, and system outputs were validated using intraclass correlation (ICC), yielding strong agreement (> 0.82) across stuttering types. The system achieved classification accuracies of 89

Automatic speech recognition for low-resource languages remains fundamentally constrained by the scarcity of labeled data and computational resources required by state-of-the-art models. We present a systematic investigation into cross-lingual continuous pretraining for low-resource languages, using Perso-Arabic languages (Persian, Arabic, and Urdu) as our primary case study. Our approach demonstrates that strategic utilization of unlabeled speech data can effectively bridge the resource gap without sacrificing recognition accuracy. We construct a 3,000-hour multilingual corpus through a scalable unlabeled data collection pipeline and employ targeted continual pretraining combined with morphologically-aware tokenization to develop a 300M parameter model that achieves performance comparable to systems 5 times larger. Our model outperforms Whisper Large v3 (1.5B parameters) on Persian and achieves competitive results on Arabic and Urdu despite using significantly fewer parameters and substantially less labeled data. These findings challenge the prevailing assumption that ASR quality scales primarily with model size, revealing instead that data relevance and strategic pretraining are more critical factors for low-resource scenarios. This work provides a practical pathway toward inclusive speech technology, enabling effective ASR for underrepresented languages without dependence on massive computational infrastructure or proprietary datasets.

Computer-Assisted Language Learning systems provide speech exaggerating at mispronounced word locations as a feedback to the L2 learners. Traditionally, expert speakers recordings (which limit the scalability) are considered for this task though there are advancements in text-to-speech (TTS) that can generate native like natural sounding speech. To address these, this work proposes two novel controllable strategies for scalable speech exaggeration. One strategy is direct speech exaggeration that incorporates the proposed label conditioned tokenization in GlowTTS. Another strategy is cascading the state-of-the-art TTS to a WORLD vocoder with proposed energy and duration modifications. A subset of Tatoeba corpus, that we annotated with prominent words, is used for experimentation. Automatic and manual assessment reveals that the exaggerated speech quality from both direct and cascaded strategy with duration modification is closer to the prominent words in the native speaker's speech.

Despite Telugu being spoken by over 80 million people, speech translation research for this morphologically rich language remains severely underexplored. We address this gap by developing a high-quality Telugu--English speech translation benchmark from 46 hours of manually verified CSTD corpus data (30h/8h/8h train/dev/test split). Our systematic comparison of cascaded versus end-to-end architectures shows that while IndicWhisper + IndicMT achieves the highest performance due to extensive Telugu-specific training data, finetuned SeamlessM4T models demonstrate remarkable competitiveness despite using significantly less Telugu-specific training data. This finding suggests that with careful hyperparameter tuning and sufficient parallel data (potentially less than 100 hours), end-to-end systems can achieve performance comparable to cascaded approaches in low-resource settings. Our metric reliability study evaluating BLEU, METEOR, ChrF++, ROUGE-L, TER, and BERTScore against human judgments reveals that traditional metrics provide better quality discrimination than BERTScore for Telugu--English translation. The work delivers three key contributions: a reproducible Telugu--English benchmark, empirical evidence of competitive end-to-end performance potential in low-resource scenarios, and practical guidance for automatic evaluation in morphologically complex language pairs.

Stuttering is recognized as a prevalent speech disorder that significantly affects individuals worldwide. Identifying and diagnosing in the early stages enhances the quality of life for individuals experiencing atypical speech patterns. Traditional methods for classifying stuttering primarily depend on subjective assessments and short-term acoustic analysis. Although helpful, these methods face limitations in accurately capturing all stutter types due to their inherent subjectivity and temporal constraints. This study uses the long-term average spectrum (LTAS) values for stutter classification derived from various filter banks such as Constant Q, Gamma-tone, and Single-frequency filter banks. It also compares these LTAS-based methods with cepstral coefficients, such as MFCC and ZTWCC. Classifiers such as SVM, LSTM, and Bi-LSTM deep networks were used to study the effectiveness of these representations in accurately discerning stuttered speech from fluent speech and reported the results.