arXiv:2511.14515v1 Announce Type: cross 
Abstract: Achieving a balance between lightweight design and high performance remains a significant challenge for speech enhancement (SE) tasks on resource-constrained devices. Existing state-of-the-art methods, such as MUSE, have established a strong baseline with only 0.51M parameters by introducing a Multi-path Enhanced Taylor (MET) transformer and Deformable Embedding (DE). However, an in-depth analysis reveals that MUSE still suffers from efficiency bottlenecks: the MET module relies on a complex "approximate-compensate" mechanism to mitigate the limitations of Taylor-expansion-based attention, while the offset calculation for deformable embedding introduces additional computational burden. This paper proposes IMSE, a systematically optimized and ultra-lightweight network. We introduce two core innovations: 1) Replacing the MET module with Amplitude-Aware Linear Attention (MALA). MALA fundamentally rectifies the "amplitude-ignoring" problem in linear attention by explicitly preserving the norm information of query vectors in the attention calculation, achieving efficient global modeling without an auxiliary compensation branch. 2) Replacing the DE module with Inception Depthwise Convolution (IDConv). IDConv borrows the Inception concept, decomposing large-kernel operations into efficient parallel branches (square, horizontal, and vertical strips), thereby capturing spectrogram features with extremely low parameter redundancy. Extensive experiments on the VoiceBank+DEMAND dataset demonstrate that, compared to the MUSE baseline, IMSE significantly reduces the parameter count by 16.8\% (from 0.513M to 0.427M) while achieving competitive performance comparable to the state-of-the-art on the PESQ metric (3.373). This study sets a new benchmark for the trade-off between model size and speech quality in ultra-lightweight speech enhancement.

تتناول الورقة المعنونة 'IMSE: تحسين الكلام الفعال القائم على U-Net باستخدام الالتفاف العمقي من Inception والانتباه الخطي المدرك للسعة' التحديات المتعلقة بتحسين الكلام (SE) على الأجهزة ذات الموارد المحدودة. تنتقد الأساليب الحالية مثل MUSE، التي على الرغم من انخفاض عدد المعلمات فيها، إلا أنها تواجه مشكلات في الكفاءة بسبب الآليات المعقدة. تقترح الشبكة IMSE المبتكرة إدخال ابتكارين رئيسيين: استبدال وحدة MET بانتباه خطي مدرك للسعة، واستخدام الالتفاف العمقي من Inception لتحسين الأداء.

El artículo titulado 'IMSE: Mejora de voz eficiente basada en U-Net utilizando convolución de profundidad de Inception y atención lineal consciente de la amplitud' aborda los desafíos de la mejora de voz (SE) en dispositivos con recursos limitados. Critica métodos existentes como MUSE, que, a pesar de su bajo número de parámetros, enfrentan problemas de eficiencia debido a mecanismos complejos. La red IMSE propuesta introduce dos innovaciones: reemplazar el módulo Multi-path Enhanced Taylor con Atención Lineal Consciente de la Amplitud y utilizar Convolución de Profundidad de Inception para me…

L'article intitulé 'IMSE : Amélioration de la parole efficace basée sur U-Net utilisant la convolution profondeur-inception et l'attention linéaire consciente de l'amplitude' aborde les défis de l'amélioration de la parole (SE) sur des appareils à ressources limitées. Il critique les méthodes existantes comme MUSE, qui, malgré leur faible nombre de paramètres, rencontrent des problèmes d'efficacité en raison de mécanismes complexes. Le réseau IMSE proposé introduit deux innovations : remplacer le module Multi-path Enhanced Taylor par l'attention linéaire consciente de l'amplitude et utiliser l…

The paper titled 'IMSE: Efficient U-Net-based Speech Enhancement using Inception Depthwise Convolution and Amplitude-Aware Linear Attention' addresses the challenges of speech enhancement (SE) on resource-constrained devices. It critiques existing methods like MUSE, which, despite their low parameter count, face efficiency issues due to complex mechanisms. The proposed IMSE network introduces two innovations: replacing the Multi-path Enhanced Taylor module with Amplitude-Aware Linear Attention, which resolves amplitude-ignoring problems, and employing Inception Depthwise Convolution for enhanc…

IMSE: Efficient U-Net-based Speech Enhancement using Inception Depthwise Convolution and Amplitude-Aware Linear Attention

arXiv:2511.21051v1 Announce Type: new 
Abstract: Images evoke emotions that profoundly influence perception, often prioritized over content. Current Image Emotional Synthesis (IES) approaches artificially separate generation and editing tasks, creating inefficiencies and limiting applications where these tasks naturally intertwine, such as therapeutic interventions or storytelling. In this work, we introduce MUSE, the first unified framework capable of both emotional generation and editing. By adopting a strategy conceptually aligned with Test-Time Scaling (TTS) that widely used in both LLM and diffusion model communities, it avoids the requirement for additional updating diffusion model and specialized emotional synthesis datasets. More specifically, MUSE addresses three key questions in emotional synthesis: (1) HOW to stably guide synthesis by leveraging an off-the-shelf emotion classifier with gradient-based optimization of emotional tokens; (2) WHEN to introduce emotional guidance by identifying the optimal timing using semantic similarity as a supervisory signal; and (3) WHICH emotion to guide synthesis through a multi-emotion loss that reduces interference from inherent and similar emotions. Experimental results show that MUSE performs favorably against all methods for both generation and editing, improving emotional accuracy and semantic diversity while maintaining an optimal balance between desired content, adherence to text prompts, and realistic emotional expression. It establishes a new paradigm for emotion synthesis.

تم تقديم MUSE، وهو إطار جديد لتوليد المشاعر في الصور، حيث يعالج عدم الكفاءة في الأساليب الحالية لتوليد المشاعر في الصور (IES) من خلال دمج مهام توليد المشاعر وتحريرها. تعتمد هذه الطريقة على تحسين الوقت الاختباري، مما يسمح بتوجيه مستقر للتوليد دون الحاجة إلى تحديثات إضافية للنموذج أو مجموعات بيانات متخصصة.

MUSE, un nuevo marco para la síntesis emocional en imágenes, ha sido introducido, abordando las ineficiencias de los métodos actuales de Síntesis Emocional de Imágenes (IES) al integrar tareas de generación y edición emocional. Este enfoque aprovecha la Escalabilidad en el Tiempo de Prueba, permitiendo una guía estable de la síntesis sin necesidad de actualizaciones adicionales del modelo o conjuntos de datos especializados.

MUSE, un nouveau cadre pour la synthèse émotionnelle dans les images, a été introduit, abordant les inefficacités des méthodes actuelles de synthèse émotionnelle d'images (IES) en intégrant les tâches de génération et d'édition émotionnelle. Cette approche s'appuie sur le Test-Time Scaling, permettant une guidance stable de la synthèse sans nécessiter de mises à jour supplémentaires du modèle ou de jeux de données spécialisés.

MUSE, a new framework for emotional synthesis in images, has been introduced, addressing inefficiencies in current Image Emotional Synthesis (IES) methods by integrating emotional generation and editing tasks. This approach leverages Test-Time Scaling, allowing for stable synthesis guidance without the need for additional model updates or specialized datasets.

IMSE: Efficient U-Net-based Speech Enhancement using Inception Depthwise Convolution and Amplitude-Aware Linear Attention

Was this article worth reading? Share it

LucidQuery AI

Voice-gen.ai

Sellm