论文解读
SALSA: Speech Aware LLM Adaptation via Learned Steering Activation Vectors
语音识别 | 7.2/10
论文速递
论文解读
SN-WER: Script-Normalized WER for Multi-Script Indic ASR Evaluation
语音识别 | 5.3/10
论文解读
SpeechEditBench: A Bilingual Multi-Attribute Benchmark for Instruction-Guided Speech Editing
语音编辑 | 8.7/10
论文解读
Spiking and Event-driven Neuromorphic Mamba Models for Efficient Speech Recognition
语音识别 | 7.5/10
论文解读
Sympatheia: Emotionally Adaptive Voice Assistant with Continuous Affect Conditioning
语音合成 | 9.6/10
论文解读
Temporally-Aligned Evaluation for Audio-Driven Talking Head Generation
语音合成 | 7.6/10
论文解读
UniVocal: Unified Speech-Singing Code-Switching Synthesis
语音合成 | 8.9/10
论文解读
WAXAL-NET: Finetuned Edge ASR Across 19 African Languages
语音识别 | 8/10
论文解读
When Tabular Foundation Models Transfer Across Modalities: A Systematic Evaluation Across 95 Datasets, 7 Modalities, and Two Regimes
音频分类 | 7.1/10
每日研究速递
语音/音乐/音频论文速递 2026-06-02
共分析 35 篇语音/AI 论文
论文解读
3DAE: Binaural Quality Assessment for Audio Novel View Synthesis with Spatial Maps and Benchmark
音频质量评估 | 6.5/10
论文解读
A Unified and Reproducible Experimentation Framework for Speech Understanding
语音识别 | 5.5/10
论文解读
AnchorSteer: Self-Discovered Concept Injection for Structure-Preserving Music Editing
音乐生成 | 8.6/10
论文解读
不可听水印为何在扩散重生面前失效：DiffErase 的黑盒去除与保真权衡
针对语音/音乐/环境声的神经音频水印，DiffErase 在不查询检测器且不知水印方案的黑盒条件下，通过梅尔频谱的中间噪声扰动与预训练扩散去噪重生实现去除，并在五种水印系统上将检测 TPR 降至近 0 的同时以 t* 控制保真度代价。
论文解读
Chatterbox-Flash: Prior-Calibrated Block Diffusion for Streaming Zero-Shot TTS
语音合成 | 10/10
论文解读
DOA: Training-Free Decoder-Only Attention Policy for Long-Form Simultaneous Translation with SpeechLLMs
语音翻译 | 7.8/10
论文解读
Escaping the Linearity Trap: Manifold Detours for Black-Box Adversarial Attacks on Singing Audio Deepfake Detection
自监督学习 | 9.7/10
论文解读
Extracting accent features in spoken Brazilian Portuguese without sociolinguistic labels
口音识别 | 8.3/10
论文解读
FiPA-SR -- FiLM-Conditioned Perceptually Informed Audio Super-Resolution
生成对抗网络 | 8.1/10
论文解读
几何主导下如何减出材料：GaMi 的跨模态相减式解耦
针对几何变化压制材料特征的问题，GaMi 用共位毫米波与声学的共享几何一致性做对齐-校准-相减解耦，并以样本间对比抑制残差，在 20 类材料上以 95.2% 的总体准确率显著优于单模态基线，代价是需双模态同步采集与多目标联合训练。