L-SpEx: Localized Target Speaker Extraction

The data configuration and simulation of L-SpEx. The code scripts will be released in the future.

Data Generation:

  1. Download LibriSpeech(dev-clean.tar.gz, test-clean.tar.gz, train-clean-100.tar.gz, train-clean-360.tar.gz) and Wham_noise(wham_noise.zip). And move the librispeech and wham_noise to ‘data_simulation/MC-Libri2Mix/spatilize_mixture/’

  2. generate the RIRs information.

python run_sample_reverb_libri.py
  1. generate the MC-Libri2Mix dataset using RIRs information.

./generate_librimix.sh YOUR_SAVE_PATH

Environments:

python: 3.8.3

Pytorch: 1.6

GitHub

View Github