Enhancing Noise Robustness of Parkinson's Disease Telemonitoring via Contrastive Feature Augmentation

Parkinson's disease (PD) is one of the most common neurodegenerative disorder. PD telemonitoring emerges as a novel assessment modality enabling self-administered at-home tests of Unified Parkinson's Disease Rating Scale (UPDRS) scores, enhancing accessibility for PD patients. However, three types of noise would occur during measurements: (1) patient-induced measurement inaccuracies, (2) environmental noise, and (3) data packet loss during transmission, resulting in higher prediction errors. To address these challenges, NoRo, a noise-robust UPDRS prediction framework is proposed. First, the original speech features are grouped into ordered bins, based on the continuous values of a selected feature, to construct contrastive pairs. Second, the contrastive pairs are employed to train a multilayer perceptron encoder for generating noise-robust features. Finally, these features are concatenated with the original features as the augmented features, which are then fed into the UPDRS prediction models. Notably, we further introduces a novel evaluation approach with customizable noise injection module, and extensive experiments show that NoRo can successfully enhance the noise robustness of UPDRS prediction across various downstream prediction models under different noisy environments.

Emotional Text-To-Speech Based on Mutual-Information-Guided Emotion-Timbre Disentanglement

Current emotional Text-To-Speech (TTS) and style transfer methods rely on reference encoders to control global style or emotion vectors, but do not capture nuanced acoustic details of the reference speech. To this end, we propose a novel emotional TTS method that enables fine-grained phoneme-level emotion embedding prediction while disentangling intrinsic attributes of the reference speech. The proposed method employs a style disentanglement method to guide two feature extractors, reducing mutual information between timbre and emotion features, and effectively separating distinct style components from the reference speech. Experimental results demonstrate that our method outperforms baseline TTS systems in generating natural and emotionally rich speech. This work highlights the potential of disentangled and fine-grained representations in advancing the quality and flexibility of emotional TTS systems.

Tenyidie Syllabification corpus creation and deep learning applications

The Tenyidie language is a low-resource language of the Tibeto-Burman family spoken by the Tenyimia Community of Nagaland in the north-eastern part of India and is considered a major language in Nagaland. It is tonal, Subject-Object-Verb, and highly agglutinative in nature. Being a low-resource language, very limited research on Natural Language Processing (NLP) has been conducted. To the best of our knowledge, no work on syllabification has been reported for this language. Among the many NLP tasks, syllabification or syllabication is an important task in which the given word syllables are identified. The contribution of this work is the creation of 10,120 syllabified Tenyidie words and the application of the Deep Learning techniques on the created corpus. In this paper, we have applied LSTM, BLSTM, BLSTM+CRF, and Encoder-decoder deep learning architectures on our created dataset. In our dataset split of 80:10:10 (train:validation:test) set, we achieved the highest accuracy of 99.21% with BLSTM model on the test set. This work will find its application in numerous other NLP applications, such as morphological analysis, part-of-speech tagging, machine translation, etc, for the Tenyidie Language. Keywords: Tenyidie; NLP; syllabification; deep learning; LSTM; BLSTM; CRF; Encoder-decoder

MOSS-Speech: Towards True Speech-to-Speech Models Without Text Guidance

Spoken dialogue systems often rely on cascaded pipelines that transcribe, process, and resynthesize speech. While effective, this design discards paralinguistic cues and limits expressivity. Recent end-to-end methods reduce latency and better preserve these cues, yet still rely on text intermediates, creating a fundamental bottleneck. We present MOSS-Speech, a true speech-to-speech large language model that directly understands and generates speech without relying on text guidance. Our approach combines a modality-based layer-splitting architecture with a frozen pre-training strategy, preserving the reasoning and knowledge of pretrained text LLMs while adding native speech capabilities. Experiments show that our model achieves state-of-the-art results in spoken question answering and delivers comparable speech-to-speech performance relative to existing text-guided systems, while still maintaining competitive text performance. By narrowing the gap between text-guided and direct speech generation, our work establishes a new paradigm for expressive and efficient end-to-end speech interaction.

EvolveCaptions: Empowering DHH Users Through Real-Time Collaborative Captioning

Automatic Speech Recognition (ASR) systems often fail to accurately transcribe speech from Deaf and Hard of Hearing (DHH) individuals, especially during real-time conversations. Existing personalization approaches typically require extensive pre-recorded data and place the burden of adaptation on the DHH speaker. We present EvolveCaptions, a real-time, collaborative ASR adaptation system that supports in-situ personalization with minimal effort. Hearing participants correct ASR errors during live conversations. Based on these corrections, the system generates short, phonetically targeted prompts for the DHH speaker to record, which are then used to fine-tune the ASR model. In a study with 12 DHH and six hearing participants, EvolveCaptions reduced Word Error Rate (WER) across all DHH users within one hour of use, using only five minutes of recording time on average. Participants described the system as intuitive, low-effort, and well-integrated into communication. These findings demonstrate the promise of collaborative, real-time ASR adaptation for more equitable communication.

Stream RAG: Instant and Accurate Spoken Dialogue Systems with Streaming Tool Usage

End-to-end speech-in speech-out dialogue systems are emerging as a powerful alternative to traditional ASR-LLM-TTS pipelines, generating more natural, expressive responses with significantly lower latency. However, these systems remain prone to hallucinations due to limited factual grounding. While text-based dialogue systems address this challenge by integrating tools such as web search and knowledge graph APIs, we introduce the first approach to extend tool use directly into speech-in speech-out systems. A key challenge is that tool integration substantially increases response latency, disrupting conversational flow. To mitigate this, we propose Streaming Retrieval-Augmented Generation (Streaming RAG), a novel framework that reduces user-perceived latency by predicting tool queries in parallel with user speech, even before the user finishes speaking. Specifically, we develop a post-training pipeline that teaches the model when to issue tool calls during ongoing speech and how to generate spoken summaries that fuse audio queries with retrieved text results, thereby improving both accuracy and responsiveness. To evaluate our approach, we construct AudioCRAG, a benchmark created by converting queries from the publicly available CRAG dataset into speech form. Experimental results demonstrate that our streaming RAG approach increases QA accuracy by up to 200% relative (from 11.1% to 34.2% absolute) and further enhances user experience by reducing tool use latency by 20%. Importantly, our streaming RAG approach is modality-agnostic and can be applied equally to typed input, paving the way for more agentic, real-time AI assistants.

LLM-Based Multi-Task Bangla Hate Speech Detection: Type, Severity, and Target

Online social media platforms are central to everyday communication and information seeking. While these platforms serve positive purposes, they also provide fertile ground for the spread of hate speech, offensive language, and bullying content targeting individuals, organizations, and communities. Such content undermines safety, participation, and equity online. Reliable detection systems are therefore needed, especially for low-resource languages where moderation tools are limited. In Bangla, prior work has contributed resources and models, but most are single-task (e.g., binary hate/offense) with limited coverage of multi-facet signals (type, severity, target). We address these gaps by introducing the first multi-task Bangla hate-speech dataset, BanglaMultiHate, one of the largest manually annotated corpus to date. Building on this resource, we conduct a comprehensive, controlled comparison spanning classical baselines, monolingual pretrained models, and LLMs under zero-shot prompting and LoRA fine-tuning. Our experiments assess LLM adaptability in a low-resource setting and reveal a consistent trend: although LoRA-tuned LLMs are competitive with BanglaBERT, culturally and linguistically grounded pretraining remains critical for robust performance. Together, our dataset and findings establish a stronger benchmark for developing culturally aligned moderation tools in low-resource contexts. For reproducibility, we will release the dataset and all related scripts.

SLAP: Learning Speaker and Health-Related Representations from Natural Language Supervision

Speech encodes paralinguistic information such as demographics, voice quality, and health. Yet no audio foundation model supports zero-shot or out-of-distribution (OOD) generalization to these tasks. We introduce SLAP (Speaker contrastive Language-Audio Pretraining), the first model aligning speech with natural language descriptions of speaker and health metadata through contrastive learning. SLAP combines a Vision Transformer audio encoder with text encoders, trained on more than 3400 hours across 9 datasets with diverse speaker annotations. We evaluated on 38 binary classification tasks spanning demographics, voice characteristics, and clinical assessments across 14 datasets in 7 languages. SLAP achieves 62.9% average F1 in zero-shot evaluation, a 48% relative improvement over CLAP (42.4%), while demonstrating strong OOD generalization to unseen languages and clinical populations. When fine-tuned with linear probing, SLAP reaches 69.3% F1 overall and achieves best-in-class performance on health tasks (57.9% F1), surpassing larger foundation models.

High-Fidelity Speech Enhancement via Discrete Audio Tokens

Recent autoregressive transformer-based speech enhancement (SE) methods have shown promising results by leveraging advanced semantic understanding and contextual modeling of speech. However, these approaches often rely on complex multi-stage pipelines and low sampling rate codecs, limiting them to narrow and task-specific speech enhancement. In this work, we introduce DAC-SE1, a simplified language model-based SE framework leveraging discrete high-resolution audio representations; DAC-SE1 preserves fine-grained acoustic details while maintaining semantic coherence. Our experiments show that DAC-SE1 surpasses state-of-the-art autoregressive SE methods on both objective perceptual metrics and in a MUSHRA human evaluation. We release our codebase and model checkpoints to support further research in scalable, unified, and high-quality speech enhancement.

MelCap: A Unified Single-Codebook Neural Codec for High-Fidelity Audio Compression

Neural audio codecs have recently emerged as powerful tools for high-quality and low-bitrate audio compression, leveraging deep generative models to learn latent representations of audio signals. However, existing approaches either rely on a single quantizer that only processes speech domain, or on multiple quantizers that are not well suited for downstream tasks. To address this issue, we propose MelCap, a unified "one-codebook-for-all" neural codec that effectively handles speech, music, and general sound. By decomposing audio reconstruction into two stages, our method preserves more acoustic details than previous single-codebook approaches, while achieving performance comparable to mainstream multi-codebook methods. In the first stage, audio is transformed into mel-spectrograms, which are compressed and quantized into compact single tokens using a 2D tokenizer. A perceptual loss is further applied to mitigate the over-smoothing artifacts observed in spectrogram reconstruction. In the second stage, a Vocoder recovers waveforms from the mel discrete tokens in a single forward pass, enabling real-time decoding. Both objective and subjective evaluations demonstrate that MelCap achieves quality on comparable to state-of-the-art multi-codebook codecs, while retaining the computational simplicity of a single-codebook design, thereby providing an effective representation for downstream tasks.