USEI TTS

有声 - adjective - voiced

---

USEI TTS is a PyQt6 interface based on the Kokoro ONNX.

After much effort and bug-fixing, it's finally ready for release!
(so many bugs, soooo many. I understand now why most open source projects are CLI based)

Features

• User-Friendly Interface: Easy to use, even for non-technical users.
  With dark and light modes.

• Fast Rendering: It is fast, very fast even if it only works on the CPU (GPU support coming soon)

• Languages:

  • American and British English
  • Japanese
  • French
  • Spanish
  • Italian
  • Hindi
  • Brazilian Portuguese
  • Mandarin Chinese

  each with several voices, male and female.

• Handles Long Texts: Breaks long texts into chunks and allows on-the-fly playback.

• No-Multi Platform Support, Yet: Unfortunately, it is currently available for Windows (I don't have an Apple computer to build and test it).

Installation

The first time you're going to run the program, it will take some time to install some bigger packages (numpy, torch) that I didn't want to include in the installer, and to download the model files, but after that, it should load pretty quickly.

Windows

WARNING

• It will probably be flagged by the Windows Defender or any antivirus that you have. That happens because it doesn't have an official signature registered. I don't have the money to pay for it (it's like 500$ a year, ridiculous)
• You might need to run it in admin mode (don't know why this happens)

Installer

1. Download the latest version from [link].
2. Follow the installation prompts.
3. Ensure Python 3.11.9 is installed and added to your PATH.
4. Enjoy!

Portable Executable

1. Download the zip file from [link].
2. Ensure Python 3.11.9 is installed before running the application.
3. Unpack it in a folder and run the executable.

As a Python Script

1. Requirements:

   • Python 3.11.9 (ensure "Add Python to PATH" is checked during installation).

2. Steps:

   • Download the zip file:

     https://github.com/drgutman/Usei/archive/refs/heads/main.zip
     

     or clone the repository:

     gh repo clone drgutman/Usei
     

   • Create a virtual environment inside the Usei folder:

     python -m venv .venv
     

   • Activate the virtual environment:

     .venv\Scripts\activate
     

   • Install the required packages:

     pip install -r requirements.txt
     

   • Run the program:

     python main.py
     

Change log:

v1.0.0

First release

v1.1.0

Interface:

• Improved loading bar animation in the splash screen (added background animation too). There's still a small inconsistency that makes it go to 100% and then back to 98% or something but I can't figure it out for the life of me.
• There was a small window/widget flickering into existence for a millisecond before the user clicks for the first time on the Replace button in the toolbar.
• Fixed the statusbar label not showing the text entirely.
• Added a new SizeGrip because the default one wasn't visible. It was there and it was working, but you couldn't see it. (yes, I had to make a custom drawn size grip with tilted gradients, it's ridiculous how many quirks you find once you start digging)
• Made the splashscreen log widget taller.

Features/Bug fixes:

• Added python check at runtime so it doesn't crash the system anymore.
• Rolled back to python 3.11.9 for compatibility with torch
• Use UV for package management.
• Reorganized the folder structure.
• Streamlined the code and added better error logging and reporting during startup.
• Moved fugashi from requirements.txt to the module_loader function.
• Included pyopenjtalk to the source code bundled_libs along with several dlls so the user doesn't have to install cmake or vs_BuildTools to run the program. Changed the debug switch in settings to "show console tab" and added a toolbar icon for show console. (there's still a small bug when you move the audio player from the bottom dock to the right bottom dock, in which the height is wrong, but I couldn't find a way to fix it no matter what I tried)
• Modified the package check to only delete and reinstall the missing packages or those with the wrong version.

Next Steps

• Find and fix any remaining bugs
• waveform/spectrogram seekbar
• Add support for more text filetypes
• Add streaming support
• Integrate podcast functionality

☒ Implement GPU/CPU switching. no. Just NO! ... I bashed my head on this for 3 or 4 days and I couldn't make it work. It has something to do with onnxruntime versions and their compatibility with the python version, and then you need vs_code build tools, and so on, and so on, and it still crashes, so I'd rather not. It's fast enough on CPU alone.

License

This project is licensed under the MIT License. See the LICENSE file for details.

Contact

For any questions or feedback, please open an issue or start a discussion or join us on Discord

Donate

If you enjoy using this program, please help me to continue developing it by donating some money
I believe in open source, but it already took me a lot of time to get it working properly and it will take a lot more to add new features.

Star History