Skip to main content
. 2026 Jun 23;8:1800552. doi: 10.3389/fdgth.2026.1800552

Table 6.

Balanced accuracy (%) on multiple levels of classification for UnispeechSAT trained with phrase with different regimes with in-domain testing.

Level Single-task Multi-task Multi-criterion
0 90.20 92.57 90.46
1 58.01 55.82 53.61
2 47.17 42.59 49.54
3 47.17 42.90 49.54
4 36.23 33.42 34.51

Bold values indicate the highest entry within each row.