TTS_dataset_creator

create dataset from list of youtube links easily

Setup repo

git clone https://github.com/m-bain/whisperX
git clone https://github.com/GregorR/rnnoise-models.git

Install ffmpeg > 5.0

for rnnnoise support for using different rnnnoise model you can change rnnnoise path in th bash script.

conda install -c conda-forge ffmpeg==5.1.0

Setup whisperx

$ git clone https://github.com/m-bain/whisperX.git
$ cd whisperX
$ pip install -e .

Speaker Diarization

To enable Speaker. Diarization, include your Hugging Face access token that you can generate from Here after the --hf_token argument and accept the user agreement for the following models: Segmentation , Voice Activity Detection (VAD) , and Speaker Diarization

update this hf_token in youtube_to_vctk.sh script

Run crawling script

add all the links in one txt file and then run youtube_to_vctk.sh links.txt out_folder_name

vctk dataset will be in output folder.