Automatic Quality Assessment of Speech-Driven Synthesized Gestures

Figure 4

(a) Loss curve to every epoch, Bi-LSTM(attention adj) represent the proposed model with algorithm adjustment in attention layer. (b)Predict scores/original scores/errors to each files in test set(baseline).(c)Predict scores/original scores/errors to each files in test set(proposed). (d)Predict scores/original scores/errors to each files in test set(algorithm adjustment proposed).

(a)

(b)

(c)

(d)