Raavgo / Cuda-TTS

TTS model

Authors

• Chory Matthias
• Dickbauer Alexander
• Langer Antonia

Overview

This project is part of our academic course, it consists of a docker container that trains a text-to-speech model using part of the M-AILABS Speech Dataset.

Setup

In order to train the model at a decent speed GPU support needs to be enabled. This is achieved by downloading and enabeling the NVIDIA Container Toolkit.

Steps

1. Setup a Debian Based OS, preferable Ubuntu 18.04 or 20.04.
2. Download and Install Ubuntu Nvidia Drivers (Restart after!)
3. Run setup.sh with root privileges

How to train the model

1. Change the parameter in train.py
2. Run docker-compose build
3. Run docker-compose up or docker-compose up -d