arXiv subjects

Man-Wai Mak

Publications and source records attributed to Man-Wai Mak.

At least 19 recordsLinked to original sources

arXiv2026

Decoupled Latent Flow Matching for Few-Step Joint Vocal-Accompaniment Separation

Generative modeling provides a flexible way to model mixture-conditioned source distributions, but iterative diffusion and flow matching models are costly for long music signals. This paper studies joint vocal-accompaniment separation through latent flow matching, where a pretrained variational autoencoder (VAE) maps mixtures and sources into a compact latent space and a flow matching model generates vocal and accompaniment latents jointly. The proposed framework decouples semantic separation from acoustic velocity prediction through a Separation Encoder and a Velocity Decoder. To reduce sampling cost, we further apply latent adversarial post-training inspired by Flow2GAN for few-step generation. Experiments show that latent adversarial refinement can improve perceptual and separation metrics under a reduced sampling budget.

cs.SD↗

arXiv2026

Learning to Prefer Reliably: Error-Augmented Emotion Preference Optimization with Calibrated Fusion

Emotion preference learning uses pairwise comparisons between candidate descriptions to align multimodal large language models (MLLMs) with human judgments of open-ended emotion descriptions and to train reward models that capture human emotional preferences. However, conventional pairwise supervision is often sparse, typically providing only a single negative description for each positive description, and therefore offers limited coverage of the diverse ways in which an emotion description can be incorrect. In particular, models may be insufficiently exposed to semantically fluent but emotionally inconsistent descriptions. Beyond this data-level limitation, relying on a single MLLM judge introduces a distinct model-level concern: its judgments can be affected by model-specific biases when interpreting fine-grained or ambiguous multimodal emotional cues. To address these limitations, we propose Error-Augmented Preference Optimization (EAPO), a framework for improving the reliability of MLLM-based emotion preference judgment at both the data and model levels. First, we construct an error-augmented dataset by generating multiple controlled and emotion-aware negative descriptions from each preferred description. We then adapt multiple independent MLLM judges to this richer supervision and aggregate their preference margins using margin-calibrated soft fusion, which maps heterogeneous margins to a common scale before aggregation. Experiments on the MER2026-EmoPrefer Challenge dataset and our error-augmented dataset demonstrate that EAPO improves emotion preference prediction and enhances the robustness of MLLM judges when evaluating fluent descriptions that conflict with the video's multimodal emotional evidence. Our code is available at https://github.com/slash1028/EAPO-EmoPrefer.

cs.MM↗

arXiv2026

Beyond Residual Connections: Manifold-Constrained Hyper-Connections for Robust Speaker Representation Learning

Residual connections are fundamental to deep speaker recogni- tion models, such as ECAPA-TDNN and ResNet. However, standard identity mapping limits information flow to a sin- gle path, constraining representation capacity. We introduce Manifold-Constrained Hyper-Connections (mHC), reformulat- ing residual paths as a multi-stream evolution where informa- tion is mixed through a doubly stochastic matrix. By employing Sinkhorn-Knopp iterations, mHC ensures energy conservation by preserving signal intensity and feature mean, which stabi- lizes gradients and mitigates signal degradation in complex net- works. We evaluate mHC by replacing standard residual con- nections in backbones including ECAPA-TDNN, ResNet-34, Res2Net, and E-Res2Net. Extensive experiments on VoxCeleb1 demonstrate that mHC connections consistently enhance per- formance across all architectures, highlighting its effectiveness for robust speaker representation learning.

cs.SD↗

arXiv2026

EII-SCL: Harnessing Emotional Inertia for Multimodal Emotion Recognition in Conversation

Multimodal emotion recognition in conversation (MERC) achieves accurate predictions by integrating multimodal and contextual information in dialogues. While current MERC approaches focus on modeling complex contextual dependencies in conversation, they often overlook the impact of contextual emotional inertia in emotion shift, leading to sub-optimal performance. To address this issue, we propose a novel Emotional Inertia-Informed Supervised Contrastive Learning module (EII-SCL) that informs the contrastive objective by constructing inertia-affected samples within temporal windows, effectively leveraging emotional inertia as a prior while enabling seamless integration with existing MERC models without requiring additional data. Extensive experiments on IEMOCAP and MELD show that our approach consistently outperforms state-of-the-art methods.

cs.MM↗

arXiv2026

EmoEUS: Uncertainty Supervision for Multimodal Emotion Recognition in Conversation

Multimodal emotion recognition in conversation (MERC) can leverage multimodal and contextual cues to boost recognition performance. However, existing fusion approaches in MERC often ignore modality-specific uncertainty across utterances caused by conflicting cues, varying noise, and missing modality-specific signals. We propose EmoEUS, an explicit uncertainty supervision framework for MERC. EmoEUS performs uncertainty-aware multimodal fusion by dynamically weighting modalities using learned variance estimates. We also introduce an explicitly supervised loss that aligns each utterance's predicted variance with the distance between the utterance's distributional representation and its emotion- and modality-specific cluster center. Experiments on IEMOCAP and MELD show that EmoEUS consistently outperforms state-of-the-art methods.

cs.MM↗

arXiv2026

UNet-Based Fusion and Exponential Moving Average Adaptation for Noise-Robust Speaker Recognition

The joint training of speech enhancement and speaker embedding networks for speaker recognition is widely adopted under noisy acoustic environments. While effective, this paradigm often fails to leverage the generalization and robustness benefits inherent in large-scale speech enhancement pre-training. Moreover, maintaining the speaker information in the denoised speech is not an explicit objective of the speech enhancement process. To address these limitations, we proposed a scalable \textbf{U}Net-based \textbf{F}usion framework (UF-EMA) that considers the noisy and enhanced speech as a multi-channel input, thereby enabling the speaker encoder to exploit speaker information effectively. In addition, an \textbf{E}xponential \textbf{M}oving \textbf{A}verage strategy is applied to a speaker encoder pre-trained on clean speech to mitigate overfitting and facilitate a smooth transition from clean to noisy conditions. Experimental results on multiple noise-contaminated test sets showcase the superiority of the proposed approach.

eess.AS↗

arXiv2026

Xi+: Uncertainty Supervision for Robust Speaker Embedding

There are various factors that can influence the performance of speaker recognition systems, such as emotion, language and other speaker-related or context-related variations. Since individual speech frames do not contribute equally to the utterance-level representation, it is essential to estimate the importance or reliability of each frame. The xi-vector model addresses this by assigning different weights to frames based on uncertainty estimation. However, its uncertainty estimation model is implicitly trained through classification loss alone and does not consider the temporal relationships between frames, which may lead to suboptimal supervision. In this paper, we propose an improved architecture, xi+. Compared to xi-vector, xi+ incorporates a temporal attention module to capture frame-level uncertainty in a context-aware manner. In addition, we introduce a novel loss function, Stochastic Variance Loss, which explicitly supervises the learning of uncertainty. Results demonstrate consistent performance improvements of about 10\% on the VoxCeleb1-O set and 11\% on the NIST SRE 2024 evaluation set.

cs.SD↗

arXiv2025

Detecting Neurocognitive Disorders through Analyses of Topic Evolution and Cross-modal Consistency in Visual-Stimulated Narratives

Early detection of neurocognitive disorders (NCDs) is crucial for timely intervention and disease management. Given that language impairments manifest early in NCD progression, visual-stimulated narrative (VSN)-based analysis offers a promising avenue for NCD detection. Current VSN-based NCD detection methods primarily focus on linguistic microstructures (e.g., lexical diversity) that are closely tied to bottom-up, stimulus-driven cognitive processes. While these features illuminate basic language abilities, the higher-order linguistic macrostructures (e.g., topic development) that may reflect top-down, concept-driven cognitive abilities remain underexplored. These macrostructural patterns are crucial for NCD detection, yet challenging to quantify due to their abstract and complex nature. To bridge this gap, we propose two novel macrostructural approaches: (1) a Dynamic Topic Model (DTM) to track topic evolution over time, and (2) a Text-Image Temporal Alignment Network (TITAN) to measure cross-modal consistency between narrative and visual stimuli. Experimental results show the effectiveness of the proposed approaches in NCD detection, with TITAN achieving superior performance across three corpora: ADReSS (F1=0.8889), ADReSSo (F1=0.8504), and CU-MARVEL-RABBIT (F1=0.7238). Feature contribution analysis reveals that macrostructural features (e.g., topic variability, topic change rate, and topic consistency) constitute the most significant contributors to the model's decision pathways, outperforming the investigated microstructural features. These findings underscore the value of macrostructural analysis for understanding linguistic-cognitive interactions associated with NCDs.

eess.AS↗

arXiv2025

EmoQ: Speech Emotion Recognition via Speech-Aware Q-Former and Large Language Model

The performance of speech emotion recognition (SER) is limited by the insufficient emotion information in unimodal systems and the feature alignment difficulties in multimodal systems. Recently, multimodal large language models (MLLMs) have made progress in SER. However, MLLMs still suffer from hallucination and misclassification problems in complex emotion reasoning. To address these problems, we propose an MLLM-based framework called EmoQ, which generates query embeddings that fuse multimodal information through an EmoQ-Former and uses multi-objective affective learning (MAL) to achieve co-optimization. The framework also provides a soft-prompt injection strategy to inject multimodal representations into the LLM. This end-to-end architecture achieves state-of-the-art performance on the IEMOCAP and MELD datasets, providing a new multimodal fusion paradigm for SER.

cs.SD↗

arXiv2025

Class Unbiasing for Generalization in Medical Diagnosis

Medical diagnosis might fail due to bias. In this work, we identified class-feature bias, which refers to models' potential reliance on features that are strongly correlated with only a subset of classes, leading to biased performance and poor generalization on other classes. We aim to train a class-unbiased model (Cls-unbias) that mitigates both class imbalance and class-feature bias simultaneously. Specifically, we propose a class-wise inequality loss which promotes equal contributions of classification loss from positive-class and negative-class samples. We propose to optimize a class-wise group distributionally robust optimization objective-a class-weighted training objective that upweights underperforming classes-to enhance the effectiveness of the inequality loss under class imbalance. Through synthetic and real-world datasets, we empirically demonstrate that class-feature bias can negatively impact model performance. Our proposed method effectively mitigates both class-feature bias and class imbalance, thereby improving the model's generalization ability.

cs.LG↗

arXiv2025

TrInk: Ink Generation with Transformer Network

In this paper, we propose TrInk, a Transformer-based model for ink generation, which effectively captures global dependencies. To better facilitate the alignment between the input text and generated stroke points, we introduce scaled positional embeddings and a Gaussian memory mask in the cross-attention module. Additionally, we design both subjective and objective evaluation pipelines to comprehensively assess the legibility and style consistency of the generated handwriting. Experiments demonstrate that our Transformer-based model achieves a 35.56\% reduction in character error rate (CER) and an 29.66% reduction in word error rate (WER) on the IAM-OnDB dataset compared to previous methods. We provide an demo page with handwriting samples from TrInk and baseline models at: https://akahello-a11y.github.io/trink-demo/

cs.CL↗

arXiv2025

Subband Architecture Aided Selective Fixed-Filter Active Noise Control

The feedforward selective fixed-filter method selects the most suitable pre-trained control filter based on the spectral features of the detected reference signal, effectively avoiding slow convergence in conventional adaptive algorithms. However, it can only handle limited types of noises, and the performance degrades when the input noise exhibits non-uniform power spectral density. To address these limitations, this paper devises a novel selective fixed-filter scheme based on a delayless subband structure. In the off-line training stage, subband control filters are pre-trained for different frequency ranges and stored in a dedicated sub-filter database. During the on-line control stage, the incoming noise is decomposed using a polyphase FFT filter bank, and a frequency-band-matching mechanism assigns each subband signal the most appropriate control filter. Subsequently, a weight stacking technique is employed to combine all subband weights into a fullband filter, enabling real-time noise suppression. Experimental results demonstrate that the proposed scheme provides fast convergence, effective noise reduction, and strong robustness in handling more complicated noisy environments.

eess.SP↗

arXiv2025

Bayesian Learning for Domain-Invariant Speaker Verification and Anti-Spoofing

The performance of automatic speaker verification (ASV) and anti-spoofing drops seriously under real-world domain mismatch conditions. The relaxed instance frequency-wise normalization (RFN), which normalizes the frequency components based on the feature statistics along the time and channel axes, is a promising approach to reducing the domain dependence in the feature maps of a speaker embedding network. We advocate that the different frequencies should receive different weights and that the weights' uncertainty due to domain shift should be accounted for. To these ends, we propose leveraging variational inference to model the posterior distribution of the weights, which results in Bayesian weighted RFN (BWRFN). This approach overcomes the limitations of fixed-weight RFN, making it more effective under domain mismatch conditions. Extensive experiments on cross-dataset ASV, cross-TTS anti-spoofing, and spoofing-robust ASV show that BWRFN is significantly better than WRFN and RFN.

eess.AS↗

arXiv2025

MoMuSE: Momentum Multi-modal Target Speaker Extraction for Real-time Scenarios with Impaired Visual Cues

Audio-visual Target Speaker Extraction (AV-TSE) aims to isolate the speech of a specific target speaker from an audio mixture using time-synchronized visual cues. In real-world scenarios, visual cues are not always available due to various impairments, which undermines the stability of AV-TSE. Despite this challenge, humans can maintain attentional momentum over time, even when the target speaker is not visible. In this paper, we introduce the Momentum Multi-modal target Speaker Extraction (MoMuSE), which retains a speaker identity momentum in memory, enabling the model to continuously track the target speaker. Designed for real-time inference, MoMuSE extracts the current speech window with guidance from both visual cues and dynamically updated speaker momentum. Experimental results demonstrate that MoMuSE exhibits significant improvement, particularly in scenarios with severe impairment of visual cues.

cs.SD↗

arXiv2024

Variational quantum classifiers via a programmable photonic microprocessor

Quantum computing holds promise across various fields, particularly with the advent of Noisy Intermediate-Scale Quantum (NISQ) devices, which can outperform classical supercomputers in specific tasks. However, challenges such as noise and limited qubit capabilities hinder its practical applications. Variational Quantum Algorithms (VQAs) offer a viable strategy to achieve quantum advantage by combining quantum and classical computing. Leveraging on VQAs, the performance of Variational Quantum Classifiers (VQCs) is competitive with many classical classifiers. This work implements a VQC using a silicon-based quantum photonic microprocessor and a classical computer, demonstrating its effectiveness in nonlinear binary and multi-classification tasks. An efficient gradient free genetic algorithm is employed for training. The VQC's performance was evaluated on three synthetic binary classification tasks with square-, circular-, and sine-shape decision boundaries and a real-world multiclass Iris dataset. The accuracies on the three binary classification tasks were 87.5%, 92.5%, and 85.0%, respectively, and 98.8% on the real world Iris dataset, highlighting the platform's potential to handle complex data patterns.

quant-ph↗

arXiv2024

On the effectiveness of enrollment speech augmentation for Target Speaker Extraction

Deep learning technologies have significantly advanced the performance of target speaker extraction (TSE) tasks. To enhance the generalization and robustness of these algorithms when training data is insufficient, data augmentation is a commonly adopted technique. Unlike typical data augmentation applied to speech mixtures, this work thoroughly investigates the effectiveness of augmenting the enrollment speech space. We found that for both pretrained and jointly optimized speaker encoders, directly augmenting the enrollment speech leads to consistent performance improvement. In addition to conventional methods such as noise and reverberation addition, we propose a novel augmentation method called self-estimated speech augmentation (SSA). Experimental results on the Libri2Mix test set show that our proposed method can achieve an improvement of up to 2.5 dB.

cs.SD↗

arXiv2024

Asymmetric Clean Segments-Guided Self-Supervised Learning for Robust Speaker Verification

Contrastive self-supervised learning (CSL) for speaker verification (SV) has drawn increasing interest recently due to its ability to exploit unlabeled data. Performing data augmentation on raw waveforms, such as adding noise or reverberation, plays a pivotal role in achieving promising results in SV. Data augmentation, however, demands meticulous calibration to ensure intact speaker-specific information, which is difficult to achieve without speaker labels. To address this issue, we introduce a novel framework by incorporating clean and augmented segments into the contrastive training pipeline. The clean segments are repurposed to pair with noisy segments to form additional positive and negative pairs. Moreover, the contrastive loss is weighted to increase the difference between the clean and augmented embeddings of different speakers. Experimental results on Voxceleb1 suggest that the proposed framework can achieve a remarkable 19% improvement over the conventional methods, and it surpasses many existing state-of-the-art techniques.

eess.AS↗

arXiv2024

VoxGenesis: Unsupervised Discovery of Latent Speaker Manifold for Speech Synthesis

Achieving nuanced and accurate emulation of human voice has been a longstanding goal in artificial intelligence. Although significant progress has been made in recent years, the mainstream of speech synthesis models still relies on supervised speaker modeling and explicit reference utterances. However, there are many aspects of human voice, such as emotion, intonation, and speaking style, for which it is hard to obtain accurate labels. In this paper, we propose VoxGenesis, a novel unsupervised speech synthesis framework that can discover a latent speaker manifold and meaningful voice editing directions without supervision. VoxGenesis is conceptually simple. Instead of mapping speech features to waveforms deterministically, VoxGenesis transforms a Gaussian distribution into speech distributions conditioned and aligned by semantic tokens. This forces the model to learn a speaker distribution disentangled from the semantic content. During the inference, sampling from the Gaussian distribution enables the creation of novel speakers with distinct characteristics. More importantly, the exploration of latent space uncovers human-interpretable directions associated with specific speaker characteristics such as gender attributes, pitch, tone, and emotion, allowing for voice editing by manipulating the latent codes along these identified directions. We conduct extensive experiments to evaluate the proposed VoxGenesis using both subjective and objective metrics, finding that it produces significantly more diverse and realistic speakers with distinct characteristics than the previous approaches. We also show that latent space manipulation produces consistent and human-identifiable effects that are not detrimental to the speech quality, which was not possible with previous approaches. Audio samples of VoxGenesis can be found at: \url{https://bit.ly/VoxGenesis}.

cs.SD↗