Skip to content

Latest commit

 

History

History
52 lines (30 loc) · 3.04 KB

README.md

File metadata and controls

52 lines (30 loc) · 3.04 KB
Raw

Firefox Plugin for using GPT-SoVITS as a screen reader


Fully local and self contained: the ONLY network traffic is to the single API endpoint that you define

You can add this plugin by one of:

  • using the signed xpi file provided
  • zipping all the files and renaming the archive from .zip to .xpi (you could set up an account at https://addons.mozilla.org and sign your own if you want to eliminate errors)
  • putting about:debugging#/runtime/this-firefox in your address bar, using "Load Temporary Addon..." and pointing it at the manifest file

Change the plugin preferences to something like https://tts.yourdomain.com/api_v2 or whatever your actual domain is and make sure you have your sovits api_v2.py api running on that url.

If you want some of the API introspection quality-of-life improvements (listbox of available voice samples, etc) use my patched api_v2.py at https://github.com/cpumaxx/GPT-SoVITS/blob/main/api_v2.py

You may need to use the -c parameter to point to a custon YAML file that defines things like your preferred ckpt and pth files.

You need at least one character+emotion defined to use this plugin.

You will need to put any sound samples you want to use on your SoVITS server, in the same folder as the api_v2.py script.

Be very careful to use a high-quality voice samples between 3-10 seconds and set the "Prompt Text" to precisely what is being said in the sample.

You can set up SoVITS and get the required files from these resources:

https://github.com/RVC-Boss/GPT-SoVITS/

https://rentry.org/GPT-SoVITS-guide

https://huggingface.co/cpumaxx/SoVITS-anime-mini-tts

Running in a Google Colab environment

You can run the API server in Google colab free tier and tunnel it out to the plugin via ngrok. Its a bit slow, but it works.

https://colab.research.google.com/github/tyc0on/GPT-SoVITS-colab/blob/main/GPT-SoVITS-colab.ipynb

First, run it once so that it can auto-download the pretrained models and you can verify that it's working.

Finally you can change the !python webui.py line to !python gpt_sovits_api_colab_ngrok.py after uploading https://pastebin.com/YfjwVTF7 into the /content folder.

Make sure to change the ngrok token placeholder in the script to your own.

You'll need to upload your .wav files, any other models and api server config yaml files and adjust the run parameters accordingly.

Manual deployment for the privacy and security sensitive

  • Clone the project with git clone https://github.com/cpumaxx/sovits-ff-plugin
  • Change the manifest.js from <all_urls> to your API endpoint and change the ID (you can't re-use them since they will be rejected my Mozilla signing)
  • zip up the manifest, all the js/html/css/svg files and change the extension from zip to xpi
  • Sign the xpi file at https://addons.mozilla.org (you will need to create an account). Use the self-host option.
  • Go to "Extensions" in your Firefox, click on the gear, icon and "Install add-on from file"