Skip to content

Latest commit

 

History

History
103 lines (83 loc) · 1.63 KB

experiments.md

File metadata and controls

103 lines (83 loc) · 1.63 KB
Raw

Experiments

Training the TDNN for 7 epochs, on 91100 utterances, 70 archives and 163 iterations.

Fusion

Evaluating:

  • 10s: on 20284 segments (20282 for fusion cases involving pitch as a single feature)

Testing:

  • 10s: on 18663 segments (18649 for fusion cases involving pitch as a single feature -- less by 0.075%)
  • 3s: on 46202 segments (45801 for fusion cases involving pitch as a single feature -- less by 0.877%)

MFCC

Training runtime: 23D features Accuracy: 0.941 C_primary: 0.078

MFCC + pitch

Training runtime: 23D features Accuracy: 0.9308 C_primary: 0.0919

MFCC + energy

Training runtime: 23D features Accuracy: 0.9387 C_primary: 0.0802

MFCC + pitch+energy

Training runtime: 23D features Accuracy: 0.9378 C_primary: 0.0822

SDC

Training runtime: 72D features Accuracy: 0.945 C_primary: 0.070

SDC + pitch+energy

Training runtime: 77D features Accuracy: 0.943 C_primary: 0.071

SDC + pitch

Training runtime: 76D features Accuracy: 0.945 C_primary: 0.072

SDC + energy

Training runtime: 76D features Accuracy: 0.945 C_primary: 0.069

MFCC+deltas

Training runtime: 69D features Accuracy: 0.944 C_primary: 0.074

MFCC+deltas+energy

Training runtime: 69D features Accuracy: 0.9454 C_primary: 0.0697

MFCC+deltas+pitch

Training runtime: 69D features Accuracy: 0.9397 C_primary: 0.0797

MFCC+deltas+pitch+energy

Training runtime: 69D features Accuracy: 0.9451 C_primary: 0.0722

Pitch+energy

Training runtime: 5D features Accuracy: 0.860 C_primary: 0.177

Pitch

Training runtime: 4D features Accuracy: 0.695 C_primary: 0.385

Energy

Training runtime: 1D features Accuracy: 0.720 C_primary: 0.359