Skip to content

Latest commit

 

History

History
48 lines (37 loc) · 1.15 KB

README.md

File metadata and controls

48 lines (37 loc) · 1.15 KB

Install dependencies

Create a conda environment:

conda env create -f ../environment.yml

Activate the environment:

conda activate Podcastmix

Download the VCTK dataset:

curl https://datashare.ed.ac.uk/download/DS_10283_2651.zip --output VCTK-Corpus.zip && \
unzip -q VCTK-Corpus.zip

Download the music

This script will take the content of ../Jamendo/metadata.json and will download the 19412 songs from the Jamendo music streaming app

python download.py

for MacOS:

CERT_PATH=$(python3 -m certifi) && \
export SSL_CERT_FILE=${CERT_PATH} && \
export REQUESTS_CA_BUNDLE=${CERT_PATH} && \
python download.py

Create the dataset

These scripts iterates through the music and speech files and will resample and copy them to the respective train, val or test directories, according to the partition of the set (80%, 10%, 10%). The metadata will be written on csv files.

python create_dataset_music.py
python create_dataset_speech.py

Check the podcast consistency

This script will check that the lines contained on the csv metadata files matches the audio files.

python check_podcastmix_consistency.py