Installation

Local Machines

I highly recommend installing TensorFlowTTS (and TensorFlow) on a designated Conda environment. I personally prefer Miniconda over Anaconda, but either one works. To begin with, follow this guide to install Conda, and then create a new Python 3.9 environment, which I will call tensorflow.

conda create -n tensorflow python=3.9
conda activate tensorflow

In the new environment, I will install TensorFlow v2.3.1 which I have found to work for training and inference later. You can install it via pip.

pip install tensorflow==2.3.1

Afterwards, clone the forked repository and install the library plus all of its requirements.

git clone https://github.com/w11wo/TensorFlowTTS.git
cd TensorFlowTTS
pip install .

Google Cloud Virtual Machines

Installing TensorFlowTTS on a Google Cloud VM is similar to installing on a local machine. To make things easier, Google has provided us with a list of pre-built VM images that comes with TensorFlow and support for GPUs. I would go for the image: Debian 10 based Deep Learning VM for TensorFlow Enterprise 2.6 with CUDA 11.0.

Because the image already has TensorFlow installed, we just need to install the main library like the steps above

git clone https://github.com/w11wo/TensorFlowTTS.git
cd TensorFlowTTS
pip install .

For some reason, there will be a bug involving Numba, which we can easily solve by upgrading NumPy to the latest version

pip install -U numpy

And also install libsndfile1 via apt

sudo apt-get install libsndfile1

Checking for a Successful Install

A way to check if your installation is correct is by importing the library through Python. We can do so through command line.

python -c "import tensorflow_tts"

If no errors are raised, then we should be good to go!