Skip to main content

Table 8 Anti-noise test of different models

From: Anchor voiceprint recognition in live streaming via RawNet-SA and gated recurrent unit

Models Vox1T-O Vox1T-N Vox1T-M Vox1T-M-S
VoiceID [45] 6.79% 9.86% 9.13% /
RawNet2 2.49% 9.03% 6.18% 6.01%
RawNet2* 2.25% 8.48% 5.80% 5.55%
RawNet-origin-SA* 2.31% 8.35% 5.75% 5.52%
RawNet-SA 2.73% 8.90% 6.15% 6.11%
  1. “*” denotes that the network is initialized with the trained RawNet2 parameters