KALDI中ivector提取过程

1. 生成 wav.scp,spk2utt,utt2spk

./local/data_prep.sh /home/yixin/kaldi/egs/clarinet/data/clarinet_audio/wav /home/yixin/kaldi/egs/clarinet/data/clarinet_audio/transcript




# 生成第三个文件 Making spk2utt files
utils/utt2spk_to_spk2utt.pl data/train/utt2spk > data/train/spk2utt
utils/utt2spk_to_spk2utt.pl data/test/utt2spk > data/test/spk2utt

2. 统一采样率为16kHz

到wav所在的文件夹 ./sampling.sh

#!/bin/bash
for x in ./*.wav
do
    b=${x##*/}
    sox $b -r 16000 -b 16 tmp_$b
    rm -rf $b
    mv tmp_$b $b
  done
echo "Converted to 16kHz"

3.计算MFCC

# Making feats.scp files
mfccdir=mfcc
# Uncomment and modify arguments in scripts below if you have any problems with data sorting
utils/validate_data_dir.sh data/train     # script for checking prepared data - here: for data/train directory
utils/fix_data_dir.sh data/train          # tool for data proper sorting if needed - here: for 

for x in train test; do
  utils/fix_data_dir.sh data/$x
  steps/make_mfcc.sh --cmd "$train_cmd" --nj 1 data/$x exp/make_mfcc/$x $mfccdir
  sid/compute_vad_decision.sh --nj 1 --cmd "$train_cmd" data/$x exp/make_mfcc/$x $mfccdir
  
done

• 得到mfcc特征文件txt

将kaldi中的特征数据读取出来，将ark二进制文件转化成.txt格式的文件

~/kaldi/src/featbin/copy-feats --binary=false ark:raw_mfcc_train.1.ark ark,t:1.txt

4. 计算ivector

# train diag ubm
sid/train_diag_ubm.sh --nj 1 --cmd "$train_cmd" --num-threads 1 \
  data/train 1024 exp/diag_ubm_1024

#train full ubm
sid/train_full_ubm.sh --nj 1 --cmd "$train_cmd" data/train \
  exp/diag_ubm_1024 exp/full_ubm_1024

#train ivector
sid/train_ivector_extractor.sh --cmd "$train_cmd --mem 10G" \
  --num-iters 5 exp/full_ubm_1024/final.ubm data/train \
  exp/extractor_1024

#extract ivector
sid/extract_ivectors.sh --cmd "$train_cmd" --nj 1 \
  exp/extractor_1024 data/train exp/ivector_train_1024