Skip to content

Results Log

Sravana Reddy edited this page Dec 4, 2015 · 51 revisions

GMMs + 52 PLPs

Averaged over all time frames

from miniClassifier.py

Results with Stratified K-Fold (non-randomized), average over 4 folds.

3 distinct languages: [AR, MA, HI]

Chance: ~47.0270% (proportion of HI)

Covariance Accuracy
Tied 46.6155
Diagonal 41.7112
Full 36.7694
Spherical 41.7471

All frames used (independently)

from gmmClassifier.py

Results with Stratified K-Fold (non-randomized), average over 4 folds.

3 distinct languages: [AR, MA, HI]

Chance: ~47.0270% (proportion of HI)

Gaussians Tied Diagonal Full Spherical
1 47.1779 42.4462 47.3096 42.0459
2 51.0738 45.8210 50.2746 44.2103
4 49.3236 ? 51.2053 ?
8 49.7261 ? 51.2090 ?

3 similar languages: [HU, PO, RU]

Chance: ~42.0732% (proportion of HU)

Gaussians Tied Diagonal Full Spherical
1 37.5627 31.9075 36.7930 36.9445
2 40.1607 30.8366 40.6198 40.2981

All languages

Results with the single split in traintestsplit

Chance: ~9.2608% (proportion of BP in test files)

Gaussians Tied Full
1 6.8237 6.6613
2 7.3111 7.2298
4 7.5548 8.2859
8 7.1487 9.2608
16 7.8798 11.4541
32 7.9610 11.7790
64 8.3672 15.5158
128 8.6109 17.5467
256 8.8546 16.5719
512 9.9650 17.8716
1024 10.8042 12.9163