diff --git a/README.md b/README.md index 1e42c0f..c51572c 100644 --- a/README.md +++ b/README.md @@ -45,16 +45,16 @@ There is a welcome screen that covers a few other bits, along with the built in - If you have NEVER run Python on your system EVER, Windows users must install C++ development tools for Python to compile Python packages. This ia a Python requirement not an AllTalk specifc requirement. Detailed information and a link to these tools can be found in the help section [**Windows & Python requirements for compiling packages**](https://github.com/erew123/alltalk_tts#-help-with-problems). ### Linux Systems -You need `libaio-dev` or `libaio-devl` (depending on your Linux flavour), otherwise DeepSpeed will fail e.g at your terminal. -- **Debian-based systems** `sudo apt install libaio-dev` -- **RPM-based systems** `sudo yum install libaio-devel` +You need to install a few bits (depending on your Linux flavour), otherwise DeepSpeed will fail and some TTS engine. At your terminal. +- **Debian-based systems** `sudo apt install libaio-dev espeak-ng ffmpeg` +- **RPM-based systems** `sudo yum install libaio-devel espeak-ng ffmpeg` As mentioned, the **atsetup.sh** install for Standalone installations **should** install deepspeed automatically. TGWUI - If however you are wanting to install in TGWUI and you can look here for a simpler way to get DeepSpeed installed here https://github.com/erew123/alltalk_tts/releases/tag/DeepSpeed-14.2-Linux, however, after 20+ hours wrestling with the conda python toolkit installations (on Linux), it looks like Nvidia may have a fault in their installation that can break 6x of the symlinks in the conda environment folder. They are reasonably easy to fix **if needed** and the script in `/system/config/fixsymlinks.sh` **may** do it for you, though I havnt tested on every OS with every eventuality etc. ### Mac Systems -I do not have a Mac system personally, so cannot test things out. In theory, all the TTS engines **should** work on Mac, however Ive been unable to write an installation routine for AllTalk. I have a loose idea of what to do. You would need to setup a miniconda for mac with Python 3.11.9 and Pytorch 2.2.1. You would need to install (brew) Espeak. I think at that point in time, once you have a working environment, you should be able to use the `requirements_standalone.txt` file with pip, though you may want to remove the nvidia cudnn bits, as mac cant use those. You would have to `conda install pytorch::faiss-cpu` and also `conda install conda-forge::ffmpeg`. The sticking point, if you wanted to use RVC is that you need a version of Fairseq that would work on Mac and Python 3.11.x. The only way I can think of doing that would be to compile a build of this https://github.com/VarunGumma/fairseq (if its possible), though I have no idea how to do that on a mac. So using RVC on Mac may be a no-no, or it may be possible for someone who wants to try figuring out Fairseq 12.2+ on mac. +I do not have a Mac system personally, so cannot test things out. In theory, all the TTS engines **should** work on Mac, however Ive been unable to write an installation routine for AllTalk. I have a loose idea of what to do. You would need to setup a miniconda for mac with Python 3.11.9 and Pytorch 2.2.1. You would need to install (brew) espeak-ng and ffmpeg (like Linux). I think at that point in time, once you have a working environment, you should be able to use the `requirements_standalone.txt` file with pip, though you may want to remove the nvidia cudnn bits, as mac cant use those. You would have to `conda install pytorch::faiss-cpu` and also `conda install conda-forge::ffmpeg`. The sticking point, if you wanted to use RVC is that you need a version of Fairseq that would work on Mac and Python 3.11.x. The only way I can think of doing that would be to compile a build of this https://github.com/VarunGumma/fairseq (if its possible), though I have no idea how to do that on a mac. So using RVC on Mac may be a no-no, or it may be possible for someone who wants to try figuring out Fairseq 12.2+ on mac. ### ALL Systems - General requirements When I say general requirements, this is what I have tested it on. If you vary outside of this setup, expect issues. I will not have time to figure out why it doesnt work on Python version x and Pytorch version x, I dont have the resource/time available.