Skip to main content
EURASIP Journal on Audio, Speech, and Music Processing

Table 5 Training and decoding time (h=hour, m=minute, s=second)

From: A hybrid input-type recurrent neural network for LVCSR language modeling

RNNLMs

W

PM

H-F

H-R1

H-R2

#Word

35 K

0 K

35 K

30 K

25 K

#PM

0 K

25 K

25 K

5 K

10 K

Training

36 h 35 m

38 h 13 m

67 h 19 m

63 h 12 m

28 h 20 m

Decoding

08 m 25 s

10 m 22 s

17 m 47 s

15 m 48 s

08 m 15 s

Back to article page