论文解读
Snore Sound Classification Based on Physiological Features and Adaptive Loss Function
音频分类 | 6.5/10
音频分类
子标签
下方只统计已核验方向的论文记录，按已核身份去重，身份待核时按页面保留。“直接”只含明确标注该方向的论文；“含子树”同时包含下级方向。点击名称进入论文库继续筛选；数量与上方按标签名称收录的页面数不同。
- 声学场景分类 直接 16 篇 · 含子树 16 篇
论文解读
Speech Emotion Recognition based on Hierarchical Transformer with Shifted Windows
语音情感识别 | 8.0/10
论文解读
Spiking Temporal-Enhanced Network for Zero-Shot Audio-Visual Learning
音频分类 | 7.0/10
论文解读
Testing The Efficient Coding Hypothesis Beyond Humans: The Auditory Kernels of Bat Vocalizations
生物声学 | 7.5/10
论文解读
Thinking While Listening: Simple Test Time Scaling for Audio Classification
音频分类 | 6.5/10
论文解读
Timbre-Aware Audio Difference Captioning for Anomalous Machine Sounds without Paired Training Data via Synthetic Perturbations
音频分类 | 7.5/10
论文解读
Timbre-Based Pretraining with Pseudo-Labels for Multi-Instrument Automatic Music Transcription
音乐信息检索 | 7.0/10
论文解读
Transfer Learning for Paediatric Sleep Apnoea Detection using Physiology-Guided Acoustic Models
音频分类 | 7.0/10
论文解读
Transferable Audio Lottery Tickets: Gradient Accumulation for Extreme Sparsity
音频分类 | 7.0/10
论文解读
UMV: A Mixture-Of-Experts Vision Transformer with Multi-Spectrogram Fusion for Underwater Ship Noise Classification
音频分类 | 7.5/10
论文解读
Unsupervised Discovery and Analysis of the Vocal Repertoires and Patterns of Select Corvid Species
生物声学 | 7.5/10
论文解读
UVT-LM: Unifying Visual and Tactile Perception with Language Model
跨模态 | 7.0/10
论文解读
WaveSpikeNet: A Wavelet-Spiking Fusion Architecture for Audio Classification on Edge Devices
音频分类 | 7.5/10
论文解读
When Audio Matters: A Lightweight, Hierarchical Fusion Model for Speech and Non-Verbal Emotion Recognition
语音情感识别 | 8.0/10
论文解读
Meta-Ensemble Learning with Diverse Data Splits for Improved Respiratory Sound Classification
音频分类 | 8.0/10
论文解读
Audio Effect Estimation with DNN-Based Prediction and Search Algorithm
音乐理解 | 8.0/10
论文解读
Deep Hierarchical Knowledge Loss for Fault Intensity Diagnosis
1. **要解决什么问题**：传统故障强度诊断方法将各类故障视为独立标签，忽略了物理状态之间固有的层次依赖关系（如“空化”是“初期空化”、“稳定空化”等的父类），这限制了模型的性能和鲁棒性。 2. **方法核心是什么**：提出一个名为DHK的通用框架，其核心是设计两个新的损失函数：**层次树损失
论文解读
Explicit Dropout: Deterministic Regularization for Transformer Architectures
这篇论文旨在解决传统Dropout方法依赖随机掩码、正则化效果不透明且难以精确控制的问题。其核心方法是提出一种确定性公式，将Dropout重新表述为一个可直接加入训练损失函数的显式正则化项，并推导出了适用于Transformer架构中注意力机制（Q、K、V）和前馈网络的正则化表达式。与已有方法相比，
论文解读
Environmental Sound Deepfake Detection Using Deep-Learning Framework
本文针对环境声音（如声音事件、声音场景）的深度伪造检测这一新兴任务，提出了一个系统的深度学习框架。**核心贡献**在于通过大量实验，系统评估了不同频谱图（MEL, CQT, Gammatone）、多种CNN架构（ResNet, Inception等）以及预训练模型（BEATs）在该任务上的表现，并验
论文解读
Incremental learning for audio classification with Hebbian Deep Neural Networks
本文针对音频分类中的增量学习（持续学习）问题，提出了一种受生物启发的解决方案。核心是解决深度学习模型在学习新任务时对旧知识的“灾难性遗忘”。作者首次将**Hebbian学习**（一种基于神经元同步激活的无监督、无反馈学习规则）与**增量学习**相结合，并设计了一个**核塑性**机制。该机制通过分析训