mrq
/
ai-voice-cloning
30
71
Fork 41
Code Issues 221 Pull Requests 5 Packages Projects Releases Wiki Activity
Page: Installation
Pages
Collecting Samples Generate History Home Installation Issues Settings Training Utilities Web UI
6 Installation
mrq edited this page 2023-07-03 01:27:53 +07:00

This software has been tested both locally on Windows and Linux, as well as Jupyter instances.

Installing

Outside of the very small prerequisites, everything needed to get TorToiSe working is included in the repo.

Pre-Requirements

Windows:

Linux:

  • python3.x (tested with 3.10)
  • git
  • ROCm for AMD, CUDA for NVIDIA
  • FFMPEG:
    • only needed when preparing datasets for training/finetuning

Bark

Bark is supported, but very experimental. Voice cloning with Bark using user-provided voices is very inconsistent, but the core Bark seems fine. To properly install it, ensure you:

  • git clone https://github.com/suno-ai/bark ./modules/bark/
  • pip3 install -e ./modules/bark/
  • pip3 install vocos

And to switch to Bark, start the web UI with "--tts-backend="bark"(or edit./config/exec.json's tts-backendentry tobark`).

A lot of the web UI's dials and knobs are removed when using Bark to generate voice clips.

VALL-E

VALL-E is supported, but extremely experimental, using my fork, and its weights with my fork are still being baked.

If you happened to source your own (or when I provide them), ensure you have installed the above package with:

  • git clone https://git.ecker.tech/mrq/vall-e ./modules/vall-e/
  • pip3 install -e ./modules/vall-e/

And to switch to VALL-E, start the web UI with "--tts-backend="vall-e"(or edit./config/exec.json's tts-backendentry tovall-e`).

A lot of the web UI's dials and knobs are removed when using VALL-E to generate voice clips.

CUDA Version

For NVIDIA cards, the setup script assumes your card support CUDA 11.7. If your GPU does not, simply edit the setup script to the right CUDA version. For example: pip install torch torchvision torchaudio --extra-index-url https://download.pytorch.org/whl/cu113 instead.

Setup

Windows

Download Python and Git and run their installers.

After installing Python, open the Start Menu and search for Command Prompt. Type cd , then drag and drop the folder you want to work in (experienced users can just cd <path> directly), then hit Enter.

Paste git clone https://git.ecker.tech/mrq/ai-voice-cloning to download TorToiSe and additional scripts, then hit Enter.

Afterwards, run the setup script, depending on your GPU, to automatically set things up.

  • AMD: setup-directml.bat
  • NVIDIA: setup-cuda.bat

If you've done everything right, you shouldn't have any errors.

Note on DirectML Support

PyTorch-DirectML is very, very experimental and is still not production quality. There's some headaches with the need for hairy kludgy patches.

These patches rely on transfering the tensor between the GPU and CPU as a hotfix, so performance is definitely harmed.

Both the conditional latent computation and the vocoder pass have to be done on the CPU entirely because of some quirks with DirectML.

On my 6800XT, VRAM usage climbs almost the entire 16GiB, so be wary if you OOM somehow. Low VRAM flags may NOT have any additional impact from the constant copying anyways.

For AMD users, I still might suggest using Linux+ROCm as it's (relatively) headache free, but I had stability problems.

Training is currently very, very improbably, due to how integrated it seems to be with CUDA. If you're fiending to train on your AMD card, please use Linux+ROCm, but I have not tested this myself.

Linux

First, make sure you have both python3.x and git installed, as well as the required compute platform according to your GPU (ROCm or CUDA).

Simply run the following block:

git clone https://git.ecker.tech/mrq/ai-voice-cloning
cd tortoise-tts
chmod +x *.sh

Then, depending on your GPU:

  • AMD: ./setup-rocm.sh
  • NVIDIA: ./setup-cuda.sh

And you should be done!

Note for AMD users

Due to the nature of ROCm, some little problems may occur. I personally have some stability issues, but they seem to only happen if I haphazardly OOM.

BitsAndBytes support requires some additional work, but most of it is automated with the setup-rocm-bnb.sh helper script.

Updating

To check for updates, simply run update.bat (or update.sh). It should pull from the repo, as well as fetch for any new dependencies.

git pull should also update its submodules, as long as I also have the latest commit pushed to this repo.

Alternative Methods

Below are some non-baremetal ways to run the software. These are either provided by other users (Docker) or are actively not supported by be due to how the underlying backend changes (notebooks). I cannot really do anything when it comes to issues popping up from the below methods, as it's beyond my support.

Docker

Notebooks

The repo provides a notebooks to run the software on an instance rather than local hardware.

Colab

Notebook: https://git.ecker.tech/mrq/ai-voice-cloning/raw/branch/master/notebook_colab.ipynb

For colab use, simply:

  • download the above notebook
  • navigate to Colab
  • upload the notebook
  • run the Installation block
  • run the Drive Mount block
  • run EITHER the "Run (Inlined)" or "Run (Non-inlined)"
  • add in your voices through Google Drive (if mounted) or upload them on the sidebar on the left

Caveats

Colab has some inconsistent problems, from restricting constant use, to randomly crashing sometimes and disconnect from the runtime, to having to gamble which GPU you do get, to constantly needing to re-initialize the entire software, to being gimped by your Drive space (if seeking persistent storage).

But it's free.

Paperspace

Notebook: https://git.ecker.tech/mrq/ai-voice-cloning/raw/branch/master/notebook_paperspace.ipynb

For paperspace use, simply:

  • download the above notebook
  • navigate to Paperspace
  • create either a new account or login to an existing one
  • create a new Gradient project
  • create a notebook
  • click start from scratch
  • select your machine (this can be changed later on)
  • select 6 hours for auto-shutdown time (this can be changed later)
  • click start notebook
  • upload the notebook file
  • run the Installation block
  • run the Running block

Caveats

Paperspace has some glaring problems that have really, really soured me from using it.

  • requires a phone number to create an account (some online SMS burners work, at least).
  • it's a different kind of gamble when trying to nab up a "free" (unmetered) GPU. While it's nice to know what you're getting, it's still throwing it to chance on whether you can nab a decent GPU (if using the paid tiers)
  • if you lose connection for a moment, or refresh the page, you lose console output from the web UI. You'll have to restart the kernel (runtime) before interacting with the notebook.
  • I've had my account deactivated because it "detected" ngrok running when I was simply using Voldy's Web UI (which uses gradio). No warning was issued, just login issues, until I contacted support with a "you get this one warning, if it happens again your account is gone", and I had to reinitialize everything after that, as if my account was completely wiped.
  • Then, after reverting to the free tier after not needing the growth plan anymore, it kept trying to charge my card (thankfully I had it as a virtual card, so it couldn't charge me), and it kept persisting until I got a "final warning" email. I logged in a few weeks or months after that to download my files, and the next day my account was trapped in limbo.

Paperspace is one of those "too good to be true" deals. While being able to use an A4000 for what amounts to $8 a month, or a A100-80G for what amounts to $40 a month, the problems that come with it, in my experinece, are definitely not worth easy compute. I definitely will not spend any more money on it out of principle of being fucked over before.

Table of Contents