kaldi 中run_ivector_common

kaldi 中run_ivector_common.sh

在使用GMM-HMM对音频和文本进行对齐之后，在使用DNN网络的时候，kaldi中会加入说话人信息，一般用ivector特征，kaldi中run_ivector_common.sh 对特征做了进一步的处理：

• 对特征进行速度上的扰动处理

# perturb the normal data to get the alignment _sp stands for speed-perturbed
utils/data/perturb_data_dir_speed_3way.sh data/${train_set} data/${train_set}_sp

结果，一般是对速度进行0.9、1.0、1.1的扰动处理

sp1.1-00000349544890655844_0-3083248710_0001 sox -t wav /data1/speechResource/run8k-train/ctrip_cancel/20190907/wavs/00000349544890655844_0/00000349544890655844_0-3083248710_0001.wav -t wav - speed 1.1 |
sp1.1-00000349544890655844_0-3083248722_0011 sox -t wav /data1/speechResource/run8k-train/ctrip_cancel/20190907/wavs/00000349544890655844_0/00000349544890655844_0-3083248722_0011.wav -t wav - speed 1.1 |

• 音量进行扰动

  # do volume-perturbation on the training data prior to extracting hires
  # features; this helps make trained nnets more invariant to test data volume.
  utils/data/perturb_data_dir_volume.sh data/${train_set}_sp_hires$online_affix || exit 1;

• 00000301430623824267_0-784002119_0004 sox --vol 0.913571714058 -t wav /data1/speechResource/run8k-train/ctrip_ivr/20190506/wavs/00000301430623824267_0/00000301430623824267_0-784002119_0004.wav -t wav - |
  00000301430623824267_0-784002122_0007 sox --vol 0.610468906799 -t wav /data1/speechResource/run8k-train/ctrip_ivr/20190506/wavs/00000301430623824267_0/00000301430623824267_0-784002122_0007.wav -t wav - |