Skip to main content

Advertisement

Table 5 Scenario 1: adaptation of the acoustic and linguistic models; results are given in terms of word accuracy (%). The baseline system (acoustic and linguistic models are trained on Ohm_N) is given in column "Ohm_N" and is identical in all three tables. "" denotes the arithmetic (unweighted) mean. The average of the four subsets weighted by the prior probabilities of the four classes is given in row "Mont."

From: On the Impact of Children's Emotional Speech on Acoustic and Language Models

  Acoustic models trained on
Test set Ohm_M Ohm_N Ohm_E Ohm_A
Mont_M 43.1 43.6 34.2 32.8
Mont_N 44.9 60.3 54.0 55.8
Mont_E 42.8 61.3 74.8 67.2
Mont_A 51.2 64.9 75.5 73.5
45.5 57.5 59.6 57.3
Mont 45.0 60.3 56.2 57.1
  Linguistic bigrams trained on
Test set Ohm_M Ohm_N Ohm_E Ohm_A
Mont_M 49.3 43.6 37.4 38.8
Mont_N 56.0 60.3 58.0 59.9
Mont_E 56.3 61.3 67.0 67.0
Mont_A 60.1 64.9 68.0 68.5
55.4 57.5 57.6 58.6
Mont 56.1 60.3 58.8 60.5
  Acoustic and linguistic models trained on
Test set Ohm_M Ohm_N Ohm_E Ohm_A
Mont_M 47.4 43.6 32.0 30.6
Mont_N 40.8 60.3 52.6 54.7
Mont_E 35.7 61.3 76.5 70.2
Mont_A 46.0 64.9 75.3 75.3
42.5 57.5 59.6 57.7
Mont 40.8 60.3 55.1 56.4