Spark-TTS

Overview

Spark-TTS builds on a Qwen-based LLM and the BiCodec single-stream token representation for efficient, high-quality speech. It supports zero-shot cloning as well as creating brand-new voices by adjusting attributes like gender, pitch and speed. The streamlined design keeps the pipeline simple.

What makes Spark-TTS special?

• Focus: LLM token-based synthesis.
• Availability: Free & Open-Source.
• Use cases: Zero-shot cloning and controllable voice creation.

Visit official site