Table 7 Performance comparison of models on noise datasets across different SNR levels on v1 dataset.
From: Dynamic convolution models for cross-frontend keyword spotting
Noise | SNR (dB) | Model | |||
|---|---|---|---|---|---|
DynTempConvNet + DML | DynTempConvNet | TENet6 | TCNet14 | ||
Urban | 20 | \(95.83\pm\)0.04 | \(95.17\pm\)0.11 | \(94.63\pm\)0.24 | \(93.90\pm\)0.13 |
15 | \(94.23\pm\)0.12 | \(93.97\pm\)0.05 | \(93.33\pm\)0.15 | \(92.90\pm\)0.21 | |
10 | \(92.70\pm\)0.06 | \(92.00\pm\)0.17 | \(90.93\pm\)0.30 | \(89.70\pm\)0.19 | |
5 | \(89.57\pm\)0.19 | \(87.90\pm\)0.11 | \(86.47\pm\)0.17 | \(84.63\pm\)0.22 | |
0 | \(82.50\pm\)0.08 | \(80.63\pm\)0.13 | \(78.60\pm\)0.35 | \(76.70\pm\)0.31 | |
WHAM | 20 | \(95.80\pm\)0.12 | \(95.10\pm\)0.12 | \(94.80\pm\)0.18 | \(94.43\pm\)0.23 |
15 | \(94.80\pm\)0.13 | \(94.10\pm\)0.09 | \(93.40\pm\)0.21 | \(91.83\pm\)0.11 | |
10 | \(91.90\pm\)0.21 | \(91.17\pm\)0.17 | \(90.30\pm\)0.31 | \(87.63\pm\)0.15 | |
5 | \(87.17\pm\)0.10 | \(86.30\pm\)0.23 | \(86.17\pm\)0.25 | \(82.10\pm\)0.27 | |
0 | \(78.43\pm\)0.10 | \(76.63\pm\)0.14 | \(75.50\pm\)0.17 | \(71.40\pm\)0.13 | |
