KokoroTestApp

✨ New: Voice styles are moved out of the library to the integrating application. The application provides now 28 different voice styles that can be used to generate speech.

A SwiftUI test application for the Kokoro TTS (Text-to-Speech) model, demonstrating high-quality faster than real-time neural speech synthesis on macOS and iOS using Apple's MLX framework.

Supported Platforms

• iOS 18.0+
• macOS 15.0+
• (Other Apple platforms may work as well)

NOTE: The application works only on iOS devices and it won't work on iOS emulator(s) because of lack of MLX support.

Installation

⚠️ Important: This repository uses Git LFS to store the large neural network model file (kokoro-v1_0.safetensors, ~600MB). You must have Git LFS installed and configured before cloning:

# Using Homebrew (macOS)
brew install git-lfs
git lfs install

After that, cloning and running the application is done as any other app:

1. Clone the repository (Git LFS will automatically download the model file):

   git clone https://github.com/yourusername/KokoroTestApp.git
   cd KokoroTestApp

2. Verify model file: Ensure the model file was downloaded correctly:

   ls -la Resources/kokoro-v1_0.safetensors
   # Should show a file of approximately 600MB

   If it failed, download the file manually:

   1. View Resources/kokoro-v1_0.safetensors
   2. On the right side of the screen, find the group of buttons: Raw, ⬇️, ✏️
   3. Click the download icon ⬇️, and save it in [your repository]/Resources directory

3. Open in Xcode:

   open KokoroTestApp.xcodeproj

4. Build and run the project in Xcode

Features

• High-Quality TTS: Leverages the Kokoro neural TTS model for natural-sounding speech synthesis
• Multiple voices: Supports different voice options
• Faster than real-time generation: Fast audio generation with performance metrics
• MLX integration: Optimized for Apple Silicon using the MLX machine learning framework

Dependencies

This project uses Swift Package Manager with the following dependencies:

• Kokoro iOS: Kokoro TTS engine
• MLX: Apple's MLX machine learning framework
• MisakiSwift: Misaki G2P library
• MLXUtilsLibrary: For reading the .npz file for voice styles

License

This project is licensed under the Apache 2.0 License - see the LICENSE file for details.