Local and CPU-Friendly High-Quality TTS (Text-to-Speech) Using Kokoro

Kokoro, a lightweight model with 82M parameters, supports various languages including English, Chinese, and Hindi, and enables very fast speech synthesis using only a CPU, without a GPU.

Key Features and Usage:

• Lightweight and Performance: Despite its small size of 82M parameters, it generates very realistic voices and operates stably even on older CPUs (i7-4770K).
• Docker-based Deployment: Using the Kokoro-FastAPI container image, you can set up a server instantly without complex configuration.
• OpenAI API Compatibility: It provides an interface compatible with OpenAI's Speech API, allowing easy integration into existing OpenAI-based applications.
• Various Voices: It supports about 50 different voices, which can be easily changed through environment variable settings.

Also mentioned is Speaches, an alternative that provides both TTS and STT (Whisper) functionality simultaneously.