Skip to main content
. Author manuscript; available in PMC: 2018 Dec 1.
Published in final edited form as: IEEE/ACM Trans Audio Speech Lang Process. 2017 Nov 23;25(12):2323–2336. doi: 10.1109/TASLP.2017.2758999

TABLE III.

PERs (%) of Baseline DNN, Articulatory Normalized DNN, LSTM, and BLSTM for Each Individual. “Norm.” Means a Combination of Procrustes Matching, fMLLR, and i-vector Methods

Speaker
ID
SD SI
Base.+
GMM
Base.+
DNN
Norm.+
DNN
Norm.+
LSTM
Norm.+
BLSTM
SPK1 69.1 57.5 44.8 42.2 36.1
SPK2 64.4 78.6 51.4 45.5 42.9
SPK3 69.7 43.8 32.7 31.8 29.1
SPK4 73.6 69 68.2 64.7 59.2
SPK5 66.9 55.7 45.6 42.6 32.5
SPK6 74.5 61.6 43.9 41.7 32.2
SPK7 73.4 65.9 68.8 67.8 59.7
SPK8 67.3 54 45.1 41.6 33.6
SPK9 73.1 64.7 43.3 42.8 30.2
SPK10 63.6 55.1 47.8 45.4 34.9
SPK11 63.7 57.2 48.7 46.1 38.4
SPK12 68.6 69.5 61.3 58.9 54.3
Mean 69.0 61.0 50.1 47.5 40.2
STD 3.9 9.1 10.7 10.6 11.2