paper
From Self-Supervised Speech Models to Mixture-of-Experts for Robust Anti-Spoofing
paperactiveprovisional
from-self-supervised-speech-models-to-mixture-of-experts-for-robust-anti-spoofing-817d69b8·1 events·first seen 2d agoAliases: From Self-Supervised Speech Models to Mixture-of-Experts for Robust Anti-Spoofing
Co-occurring entities
More like this (12)
Sparse Mixture-of-ExpertsSpeaker Group Encoding in Self-supervised Speech Recognition ModelsRAT: Reference-Augmented Training for ASV Anti-SpoofingTying the Loop -- Tied Expert Layers in Mixture-of-Experts Language ModelsLeveraging Audio-LLMs to Filter Speech-to-Speech Training DataCross-Modal Masking for Robust Silent Speech Synthesis Using sEMG and LipreadingBeyond Fully Random Masking: Attention-Guided Denoising and Optimization for Diffusion Language ModelsMixture of ExpertsExploring Adversarial Robustness and Safety Alignment in Multilingual Multi-Modal Large Language ModelsFrom Observation to Intervention: A Causal Audit of Expert Importance in Mixture-of-Experts ModelsWhat Do Deepfake Speech Detectors Actually Hear?Ethical and Technical Limits of Deepfake Speech Datasets
Recent events (1)
MoE architecture improves self-supervised speech model robustness for anti-spoofing
Researchers propose converting a self-supervised speech representation model into a Mixture-of-Experts (MoE) architecture to improve generalization in synthetic speech detection. Feed-forward blocks in selected encoder layers are replaced by expert networks with a layer-wise gating mechanism, allowing complementary acoustic pattern capture while preserving pretrained representations. Evaluated across 14 spoofing datasets, the approach reduces macro Equal Error Rate from 5.46% to 4.81%, an 11.9% relative improvement over the baseline.