Installation¶

In order to use EveryVoice on GPUs, you must install PyTorch and Cuda, Python 3.10 or more recent, a number of other dependencies, and EveryVoice itself. The following sections describe various ways to accomplish this.

Choosing your package manager¶

Before you start, you must choose a package manager.

We recommend using either conda or uv.

conda: somewhat slow, but highly reliable — can install all the EveryVoice dependencies, including the non-Python ones. We provide a fully automated script to do it all with conda.

You can install conda via Miniforge for free or, if you have a valid Anaconda license or can get one, Miniconda or Conda.
uv: much faster, but manages only the Python dependencies. Install uv.

If you are able to install sox, libsox-dev and ffmpeg using your OS packages or by other means (see README.md), you might prefer using uv; otherwise, use conda.

Choosing your Python version¶

EveryVoice supports Python 3.10, 3.11 and 3.12. We use 3.12 everywhere in this document, but you can use your preferred version among these instead.

Installation from PyPI¶

Direct installation from PyPI is fairly reliable.

Follow the instructions in README.md to install sox, ffmpeg, torch and torchaudio.
Ideally create a virtual environment for your project.
Install EveryVoice:

Using condaUsing uv

pip install everyvoice

uv pip install everyvoice

Installation from source¶

Clone the repo¶

You can obtain the source code by cloning the EveryVoice repo and its submodules:

git clone https://github.com/EveryVoiceTTS/EveryVoice.git
cd EveryVoice
git submodule update --init

Option 1 — Scripted installation — recommended¶

While the EveryVoice installation process has gotten simpler as the project matures, we maintain a script to automate the process and keep it reliable.

Using condaUsing uv

Run our automated environment creation script:
```
./make-everyvoice-env --name EveryVoice
conda activate EveryVoice
```
Add the option --cuda CUDA_VERSION if you need to override the default CUDA version, or --cpu to use Torch compiled for CPU use only.

Having installed sox, libsox-dev and ffmpeg (see above), run our automated environment creation script:
```
./make-everyvoice-env --uv
source .venv/bin/activate
```
Add the option --cuda CUDA_VERSION if you need to override the default CUDA version, or --cpu to use Torch compiled for CPU use only.

Option 2-a — Manual installation — Compact summary¶

Using condaUsing uv

conda create --name EveryVoice python=3.12 ffmpeg
conda activate EveryVoice
conda install sox -c conda-forge
CUDA_TAG=cu121 pip install -r requirements.torch.txt --find-links https://download.pytorch.org/whl/torch_stable.html
pip install -e '.[dev]'

Install sox, libsox-dev and ffmpeg (see above), then run:

uv venv -p 3.12 .venv
source .venv/bin/activate
uv pip install torch==2.3.1+cu121 torchaudio==2.3.1+cu121 --find-links https://download.pytorch.org/whl/torch_stable.html
uv pip install -e '.[dev]'

Option 2-b — Manual installation — Detailed¶

If you prefer to do the complete installation process manually, or if the automated process does not work for you, follow these steps.

Create the environment¶

Create a new virtual environment and activate it:

Using condaUsing uv

conda create --name EveryVoice python=3.12
conda activate EveryVoice

uv venv -p 3.12
source .venv/bin/activate

Pytorch dependencies¶

Using condaUsing uv

Install our pytorch requirements from requirements.torch.txt:

CUDA_TAG=cu121 pip install -r requirements.torch.txt --find-links https://download.pytorch.org/whl/torch_stable.html

Install our pytorch requirements specified in requirements.torch.txt, but manually. (Unfortunately, uv does not support the environment variable we use with pip and conda.)

uv pip install torch==2.3.1+cu121 torchaudio==2.3.1+cu121 --find-links https://download.pytorch.org/whl/torch_stable.html

Replace cu121 above (for CUDA 12.1) by your actual CUDA version tag (cu118 or cu121), or by cpu for a CPU-only installation.

Alternatively, you can follow the PyTorch installation instructions relevant to your hardware. Make sure you specify the version declared in requirements.torch.txt, which is 2.3.1 at the moment, if you install EveryVoice from GitHub, but 2.1.0 if you install it from PyPI.

Non-Python dependencies¶

Using condaUsing uv

Install sox and ffmpeg with conda if you didn't install them using OS packages:

conda install sox -c conda-forge
conda install ffmpeg

Uv does not provide a mechanism for installing non-Python dependencies. You must follow the sox, libsox-dev and ffmpeg installation instructions in our README.md if you want to use uv.

Handling running out of temp disk space¶

Installation will require a fair bit of space on ~/.cache and/or ~/.local, as well as your $TMPDIR (/tmp by default, if $TMPDIR is not set). If you get the error OSError: [Errno 28] No space left on device during installation, you may need to do one or more of these operations:

export TMPDIR=/path/to/a/large/tmp/space (or maybe export TMPDIR=.)
mkdir /path/to/a/large/filesystem/.cache; ln -s /path/to/a/large/filesystem/.cache ~/.cache
mkdir /path/to/a/large/filesystem/.local; ln -s /path/to/a/large/filesystem/.local ~/.local

Install EveryVoice itself¶

Install EveryVoice locally from your cloned sandbox:

Using condaUsing uv

pip install -e .

uv pip install -e .

Dev dependencies¶

Before you can run the test suites, you'll also need to install the dev dependencies:

Using condaUsing uv

pip install -e '.[dev]'

uv pip install -e '.[dev]'

Git hooks¶

If you plan to contribute to the project, please install our Git hooks:

pre-commit install
gitlint install-hook
git submodule foreach 'pre-commit install'
git submodule foreach 'gitlint install-hook'