Skip to main content

Table 10 Regression results after fusing with text classification. Turkish database was used since transcriptions are not available in the required for the other databases. Baseline acoustic system predictions are used in base-fusion. Bold results show cases where the improvement is significant compared to the baseline case but not to the base-fusion case. In the underlined bold case, improvement is significant both compared to the baseline system and the base-fusion system

From: Depression-level assessment from multi-lingual conversational speech data using acoustic and text features

Dim Baseline Base-fusion Fusion Tr+Ger Fusion Tr+Eng Fusion
    Nmin=3 Nmin=3 15 Clus.
3 12.30 10.71 9.76 9.43 9.75
4 12.45 10.68 9.54 8.88 9.74
5 12.56 10.69 9.61 8.30 9.78
10 12.45 10.91 9.83 9.74 10.05
15 12.08 10.44 9.63 10.12 9.95
20 12.87 10.91 9.65 9.62 9.66
40 13.28 11.33 10.45 9.55 10.88
80 11.58 10.12 10.67 9.37 9.99
100 11.75 10.25 10.77 9.45 10.24
200 11.32 10.03 9.98 9.76 10.40
400 11.42 10.29 10.00 9.60 9.78
800 11.31 10.31 10.19 9.79 9.88