In this project, we implement three GANs model for data augmentation on speaker recognition.

You need to download NIST 2014 i-vector Machine Learning Challenge. Convert the format of data into numpy array and put it to NIST_npy/.

The dataset.py is used to handle the dataset (ex. next_batch, sort, add), If everything is ready, execute main.py to start training.

After the training is completed, you can run generate.ipynb to restore model and generate data.

• Ubuntu 16.04 64bit
• Python 2.7.12
• Tensorflow 1.1

• Intel i7-4930K @3.40GHz
• DDR3-1600 8GB*8  - Nvidia GeForce GTX 980 3GB