Joint Enhancement and Classification using Coupled Diffusion Models of Signals and Logits

Robust classification in noisy environments remains a fundamental challenge in machine learning. Standard approaches typically treat signal enhancement and classification as separate, sequential stages: first enhancing the signal and then applying a classifier. This approach fails to leverage the semantic information in the classifier's output during denoising. In this work, we propose a general, domain-agnostic framework that integrates two interacting diffusion models: one operating on the input signal and the other on the classifier's output logits, without requiring any retraining or fine-tuning of the classifier. This coupled formulation enables mutual guidance, where the enhancing signal refines the class estimation and, conversely, the evolving class logits guide the signal reconstruction towards discriminative regions of the manifold. We introduce three strategies to effectively model the joint distribution of the input and the logit. We evaluated our joint enhancement method for image classification and automatic speech recognition. The proposed framework surpasses traditional sequential enhancement baselines, delivering robust and flexible improvements in classification accuracy under diverse noise conditions.

What Do Neurons Listen To? A Neuron-level Dissection of a General-purpose Audio Model

In this paper, we analyze the internal representations of a general-purpose audio self-supervised learning (SSL) model from a neuron-level perspective. Despite their strong empirical performance as feature extractors, the internal mechanisms underlying the robust generalization of SSL audio models remain unclear. Drawing on the framework of mechanistic interpretability, we identify and examine class-specific neurons by analyzing conditional activation patterns across diverse tasks. Our analysis reveals that SSL models foster the emergence of class-specific neurons that provide extensive coverage across novel task classes. These neurons exhibit shared responses across different semantic categories and acoustic similarities, such as speech attributes and musical pitch. We also confirm that these neurons have a functional impact on classification performance. To our knowledge, this is the first systematic neuron-level analysis of a general-purpose audio SSL model, providing new insights into its internal representation.

Clinically Inspired Symptom-Guided Depression Detection from Emotion-Aware Speech Representations

Depression manifests through a diverse set of symptoms such as sleep disturbance, loss of interest, and concentration difficulties. However, most existing works treat depression prediction either as a binary label or an overall severity score without explicitly modeling symptom-specific information. This limits their ability to provide symptom-level analysis relevant to clinical screening. To address this, we propose a symptom-specific and clinically inspired framework for depression severity estimation from speech. Our approach uses a symptom-guided cross-attention mechanism that aligns PHQ-8 questionnaire items with emotion-aware speech representations to identify which segments of a participant's speech are more important to each symptom. To account for differences in how symptoms are expressed over time, we introduce a learnable symptom-specific parameter that adaptively controls the sharpness of attention distributions. Our results on EDAIC, a standard clinical-style dataset, demonstrate improved performance outperforming prior works. Further, analyzing the attention distributions showed that higher attention is assigned to utterances containing cues related to multiple depressive symptoms, highlighting the interpretability of our approach. These findings outline the importance of symptom-guided and emotion-aware modeling for speech-based depression screening.

Under-resourced studies of under-resourced languages: lemmatization and POS-tagging with LLM annotators for historical Armenian, Georgian, Greek and Syriac

Low-resource languages pose persistent challenges for Natural Language Processing tasks such as lemmatization and part-of-speech (POS) tagging. This paper investigates the capacity of recent large language models (LLMs), including GPT-4 variants and open-weight Mistral models, to address these tasks in few-shot and zero-shot settings for four historically and linguistically diverse under-resourced languages: Ancient Greek, Classical Armenian, Old Georgian, and Syriac. Using a novel benchmark comprising aligned training and out-of-domain test corpora, we evaluate the performance of foundation models across lemmatization and POS-tagging, and compare them with PIE, a task-specific RNN baseline. Our results demonstrate that LLMs, even without fine-tuning, achieve competitive or superior performance in POS-tagging and lemmatization across most languages in few-shot settings. Significant challenges persist for languages characterized by complex morphology and non-Latin scripts, but we demonstrate that LLMs are a credible and relevant option for initiating linguistic annotation tasks in the absence of data, serving as an effective aid for annotation.

ZeroSyl: Simple Zero-Resource Syllable Tokenization for Spoken Language Modeling

Pure speech language models aim to learn language directly from raw audio without textual resources. A key challenge is that discrete tokens from self-supervised speech encoders result in excessively long sequences, motivating recent work on syllable-like units. However, methods like Sylber and SyllableLM rely on intricate multi-stage training pipelines. We propose ZeroSyl, a simple training-free method to extract syllable boundaries and embeddings directly from a frozen WavLM model. Using L2 norms of features in WavLM's intermediate layers, ZeroSyl achieves competitive syllable segmentation performance. The resulting segments are mean-pooled, discretized using K-means, and used to train a language model. ZeroSyl outperforms prior syllabic tokenizers across lexical, syntactic, and narrative benchmarks. Scaling experiments show that while finer-grained units are beneficial for lexical tasks, our discovered syllabic units exhibit better scaling behavior for syntactic modeling.

MAEB: Massive Audio Embedding Benchmark

We introduce the Massive Audio Embedding Benchmark (MAEB), a large-scale benchmark covering 30 tasks across speech, music, environmental sounds, and cross-modal audio-text reasoning in 100+ languages. We evaluate 50+ models and find that no single model dominates across all tasks: contrastive audio-text models excel at environmental sound classification (e.g., ESC50) but score near random on multilingual speech tasks (e.g., SIB-FLEURS), while speech-pretrained models show the opposite pattern. Clustering remains challenging for all models, with even the best-performing model achieving only modest results. We observe that models excelling on acoustic understanding often perform poorly on linguistic tasks, and vice versa. We also show that the performance of audio encoders on MAEB correlates highly with their performance when used in audio large language models. MAEB is derived from MAEB+, a collection of 98 tasks. MAEB is designed to maintain task diversity while reducing evaluation cost, and it integrates into the MTEB ecosystem for unified evaluation across text, image, and audio modalities. We release MAEB and all 98 tasks along with code and a leaderboard at https://github.com/embeddings-benchmark/mteb.

Disentangling Pitch and Creak for Speaker Identity Preservation in Speech Synthesis

We introduce a system capable of faithfully modifying the perceptual voice quality of creak while preserving the speaker's perceived identity. While it is well known that high creak probability is typically correlated with low pitch, it is important to note that this is a property observed on a population of speakers but does not necessarily hold across all situations. Disentanglement of pitch from creak is achieved by augmentation of the training dataset of a speech synthesis system with a speaker manipulation block based on conditional continuous normalizing flow. The experiments show greatly improved speaker verification performance over a range of creak manipulation strengths.

Data Augmentation for Pathological Speech Enhancement

The performance of state-of-the-art speech enhancement (SE) models considerably degrades for pathological speech due to atypical acoustic characteristics and limited data availability. This paper systematically investigates data augmentation (DA) strategies to improve SE performance for pathological speakers, evaluating both predictive and generative SE models. We examine three DA categories, i.e., transformative, generative, and noise augmentation, assessing their impact with objective SE metrics. Experimental results show that noise augmentation consistently delivers the largest and most robust gains, transformative augmentations provide moderate improvements, while generative augmentation yields limited benefits and can harm performance as the amount of synthetic data increases. Furthermore, we show that the effectiveness of DA varies depending on the SE model, with DA being more beneficial for predictive SE models. While our results demonstrate that DA improves SE performance for pathological speakers, a performance gap between neurotypical and pathological speech persists, highlighting the need for future research on targeted DA strategies for pathological speech.

Breaking Data Efficiency Dilemma: A Federated and Augmented Learning Framework For Alzheimer's Disease Detection via Speech

Early diagnosis of Alzheimer's Disease (AD) is crucial for delaying its progression. While AI-based speech detection is non-invasive and cost-effective, it faces a critical data efficiency dilemma due to medical data scarcity and privacy barriers. Therefore, we propose FAL-AD, a novel framework that synergistically integrates federated learning with data augmentation to systematically optimize data efficiency. Our approach delivers three key breakthroughs: First, absolute efficiency improvement through voice conversion-based augmentation, which generates diverse pathological speech samples via cross-category voice-content recombination. Second, collaborative efficiency breakthrough via an adaptive federated learning paradigm, maximizing cross-institutional benefits under privacy constraints. Finally, representational efficiency optimization by an attentive cross-modal fusion model, which achieves fine-grained word-level alignment and acoustic-textual interaction. Evaluated on ADReSSo, FAL-AD achieves a state-of-the-art multi-modal accuracy of 91.52%, outperforming all centralized baselines and demonstrating a practical solution to the data efficiency dilemma. Our source code is publicly available at https://github.com/smileix/fal-ad.

SA-SSL-MOS: Self-supervised Learning MOS Prediction with Spectral Augmentation for Generalized Multi-Rate Speech Assessment

Designing a speech quality assessment (SQA) system for estimating mean-opinion-score (MOS) of multi-rate speech with varying sampling frequency (16-48 kHz) is a challenging task. The challenge arises due to the limited availability of a MOS-labeled training dataset comprising multi-rate speech samples. While self-supervised learning (SSL) models have been widely adopted in SQA to boost performance, a key limitation is that they are pretrained on 16 kHz speech and therefore discard high-frequency information present in higher sampling rates. To address this issue, we propose a spectrogram-augmented SSL method that incorporates high-frequency features (up to 48 kHz sampling rate) through a parallel-branch architecture. We further introduce a two-step training scheme: the model is first pre-trained on a large 48 kHz dataset and then fine-tuned on a smaller multi-rate dataset. Experimental results show that leveraging high-frequency information overlooked by SSL features is crucial for accurate multi-rate SQA, and that the proposed two-step training substantially improves generalization when multi-rate data is limited.