Tortoise text-to-speech

[Decayer]

There is a pretty good, humanly-sounding text-to-speech program called Tortoise TTS.

It is free and can be installed on your machine (requires Nvidia GPU), but there is also a Google Collab page too.

It is not a real-time TTS, it takes a couple of minutes to generate a sentence, but the result is usually pretty good.

You can feed it some audio samples of your custom  voice and it will imitate it. I don't think it always does a good job at cloning the voice, but it can sound pretty close.

Each time you generate a text prompt it gives a slightly different result in terms of tone and expression.

Project GitHub page https://github.com/jnordberg/tortoise-tts

It has all the links and installation instructions. If you can't wrap your head around the instructions, here is a more precise instructions I wrote https://pastebin.com/p2RbKRWF

Link to the Collab page: https://colab.research.google.com/drive/1wVVqUPqwiDBUVeWWOUNglpGhU3hg_cbR?usp=sharing

If you don't know how to use Collab page, ask and I will explain.

GitHub page above has a link to the demos, but I can generate something for you on request.

Here is an example with the voice of female protagonist from Fallout 4:

Sometimes, you will get distortion, like weird pitch spikes. Or you can get a wrong inflection or tone. In that case you will have to try again. If it keeps happening I'll suggest rephrasing the text input.

The advantage of the local installation over Collab  is that you get 3 results per generation instead of just one, so it is more likely you will get a right result quicker.

I think it can be a good resource for modders.

Here I gathered a bunch of voice samples which you can use with this program (some voices from Fallout 4) https://www.dropbox.com/s/y6wu8ajblt69fcx/VoiceFolders.rar?dl=0

Edited August 8, 2022 by Decayer