Skip to main content

Table 15 The decoding results and word error rates of different models

From: Advanced recurrent network-based hybrid acoustic models for low resource speech recognition

Model Text (ins/del/sub percentage)
Reference aa ilap’arak’e ravi biC’o sosom rom mankana daamt’vria da gamopKizlda aKla ra Kasiatzea
DNN aa ilap’arak’e Kasiatzea (0/77/0)
LSTM aa ilap’arak’e mankana aKla ra Kasiatzea (0/54/0)
LW-BLSTM aa ilap’arak’e davT’ert masesKeo mankana daadgeba gamopKizlda aKla ra Kasiatzea (0/23/23)
LW-BGRU aa ilap’arak’e mankana daamt’vria da gamopKizlda aKla ra Kasiatzea (0/31/0)
LW-BrGRU aa ilap’arak’e ravi mankana daamt’vria da gamopKizlda aKla ra Kasiatzea (0/23/0)
  1. The id of the utterance is 81424-B-20141123-000421-030633 for Georgian
\