Skip to content

Latest commit

 

History

History
31 lines (24 loc) · 864 Bytes

README.md

File metadata and controls

31 lines (24 loc) · 864 Bytes

sound-of-laughter

L’éclat du rire (The Sound of Laughter) - AI techniques for detecting and generating laughter

Inspired by MSGGAN.

This repository contains:

  • A MSGGAN designed to work on mel spectrograms.
  • A simple dataloader designed to load data from a folder of .wav files.
  • A training script for the MSGGAN.

Training MSGGAN

If required, install the dependencies (in a venv with and upgraded pip)

$ pip install torch==1.7.1+cu110 torchvision==0.8.2+cu110 torchaudio==0.7.2 -f https://download.pytorch.org/whl/torch_stable.html
$ pip install -r requirements.txt

Set the path to the base dir of the project

$ export PYTHONPATH=$PYTHONPATH:`pwd`

Run the training script

$ python msggan/train_msggan.py \
--data_path <data_root> \
--root_path <output_path> \
--epochs 500