Table 5 Comparison of speaker trait estimation performance across prosodic features and anonymization methods.
From: Privacy-aware speaker trait and multimodal features relationship analysis in job interviews
Anon. method | Loudness46 | Timbral47 | IS12-Trait3 | |||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Ex | Co | Ag | Hr | Ex | Co | Ag | Hr | Ex | Co | Ag | Hr | Ex | Co | Ag | Hr | |
Original | 0.794 | 0.836 | 0.820 | 0.655 | 0.800 | 0.844 | 0.810 | 0.638 | 0.795 | 0.850 | 0.819 | 0.665 | 0.783 | 0.818 | 0.791 | 0.681 |
MAC29 | 0.794 | 0.832 | 0.811 | 0.636 | 0.804 | 0.838 | 0.819 | 0.634 | 0.803 | 0.835 | 0.816 | 0.661 | 0.784 | 0.823 | 0.801 | 0.663 |
PVT30 | 0.805 | 0.836 | 0.821 | 0.642 | 0.803 | 0.842 | 0.817 | 0.637 | 0.800 | 0.836 | 0.822 | 0.662 | 0.788 | 0.808 | 0.803 | 0.681 |
NAC32 | 0.786 | 0.794 | 0.783 | 0.609 | 0.777 | 0.790 | 0.781 | 0.638 | 0.776 | 0.789 | 0.777 | 0.629 | 0.773 | 0.798 | 0.780 | 0.692 |