Skip to main content

Advertisement

Table 14 Multi-condition (combined) NN training datasets for the experiments using real reverberant data

From: Single-channel dereverberation by feature mapping using cascade neural networks for robust distant speaker identification and speech recognition

Dataset name Speaker number Utterances num. per spk. per env. (pairs of utts.) Total duration of utterances (seconds)
5s.20u 5 1 54
5s.40u 5 2 108
5s.60u 5 3 162
5s.80u 5 4 213
1s.20u 1 5 70
1s.40u 1 10 138
  1. The total duration of utterances is after removing the silence parts in the beginning and ending of each recording. For the one-speaker datasets (‘1s’), the total duration is the average from five speakers’ datasets.