Skip to content

Latest commit

 

History

History
46 lines (41 loc) · 2.24 KB

using_vctk_dataset.md

File metadata and controls

46 lines (41 loc) · 2.24 KB
Raw

Getting a usable single speaker dataset from the VCTK dataset

Quick instructions

  1. download_vctk_dataset.sh Download and unzip dataset. Note that it is quite large (over 11GB)
  2. single_voice_from_VCTK_dataset.sh pNNN Repackages a single speaker's voice from this dataset and generates metadata.csv
  3. batch_resample_and_convert --dataset_dir pNNN_original_speaker --sampling_rate 16000 Batch converts flac files to wav and downsamples them. Original files are backed up in flac directory prior to conversion.

Background on using VCTK dataset for training

  • the VCTK dataset can be found here: https://datashare.ed.ac.uk/handle/10283/3443
  • Since it is a multi-speaker dataset, it can't be used directly. However, since it contains a wide variety of voices speaking the same sentences, it can be useful for finding a voice that is a close match for the target voice
  • speaker-info.txt contains details about the gender and regional dialect of each speaker in the corpus.
ID  AGE  GENDER  ACCENTS  REGION COMMENTS 
p225  23  F    English    Southern  England
p226  22  M    English    Surrey
p227  38  M    English    Cumbria
p228  22  F    English    Southern  England
p229  23  F    English    Southern  England
  • Voice samples are found in wav48_silence_trimmed directory, and are grouped in folders with each speaker's ID eg. p225
  • Transcripts for each utterance are found it the "txt" directory, also grouped in folders with each speaker's ID.
  • After I knew which voice I wanted, I did as follows:
mkdir singlevoice
cp wav48_silence_trimmed/p225/* singlevoice/wav
cp txt/p225/* singlevoice/txt
  • Each audio sample in the VCTK dataset was recorded with 2 different mics. I removed all the files recorded with the second mic to make the next step easier
rm singlevoice/*mic2
  • Finally, I used the following script to create a metadata.csv file from the file names and individual text files:
ls txt/*/*.txt | while read TXT ; do \
      BASE=`basename $TXT .txt` \
      LINE=`cat $TXT` ; \
      SPKR=`echo $BASE | awk -F_ '{print $1}'` \
      if [ $SPKR == p225 ] ; then \
         LINE=${LINE:1:-1} ; ## remove initial and final " in p225 data \
      fi \
      echo "${BASE}_mic1_output|0|$LINE" \
done >> metadata.csv