This repo has been retired and is no longer being worked on. I recommend F5-TTS as a suitable alternative. If you wish to use the code below, you can stil fork the repo/download the files locally. This repository is no longer updated or maintained - the code should not be used in production settings until it has been sanity checked for any issues.

MARS5 Text-To-Speech (TTS) Deep Voice Clone

This is a Python implementation for the MARS5 TTS repo that allows you to clone a voice with a command line interface.

Try the online demo here for a quickstart or follow the instructions below to run this project locally/offline

Roadmap

• EZ front end
• Enable PyTorch CUDA (supported devices only)

Installation

Install Python

Clone directory/download .zip file

Open CMD/Powershell in directory and run

pip install -r requirements.txt

Usage

Visual Studio Code

Download/clone this repo and open the folder in Visual Studio Code

Find your Terminal and run

python tts.py # where tts.py is the file name of the Python script

You'll be taken through the steps to start cloning - fill in your own values or press ENTER for default.

Command Line/PowerShell

Download/clone this repo and open Command Line/PowerShell from the folder (File -> Open PowerShell Window)

There's no impact on us whether we use Command Line or PowerShell. They look the same in terms of input/output.

python tts.py

Notes

• Keep reference voice between 6-10 seconds
• Tune config to get optimal output
• Provide transcript when possible
• Use sample provided for ease

Known issues

• System won't lock up but CPU usage remains at 100% throughout processing stage
• Results will vary - this is not a one shot model but seems far more impressive than ElevenLabs, Speechify at the cost of processing time