licyk/qwen-tts-webui

What it solves

This project provides a user-friendly interface and API for the Qwen3-TTS text-to-speech models. It simplifies the process of generating high-quality speech from text, including features like voice cloning, voice design, and streaming audio output, without requiring deep technical knowledge of the underlying model architecture.

How it works

It wraps the Qwen3-TTS models in a Gradio-based web interface and an OpenAI-compatible API server. The project includes optimized inference code that replaces the standard HuggingFace implementation, utilizing techniques like static KV caching, CUDA Graphs for NVIDIA GPUs, and batch processing to significantly speed up audio generation. It automatically handles model downloading and provides a local server that can be accessed via standard OpenAI SDKs.

Who it’s for

This tool is for developers and users who want to integrate Qwen3-TTS into their applications or use it locally for speech synthesis tasks. It is particularly useful for those who prefer using familiar OpenAI-style API patterns or need a graphical interface to experiment with different voices and voice cloning capabilities.

Highlights

• OpenAI API Compatibility: Supports standard OpenAI Audio Speech API endpoints, allowing for easy integration with existing tools and SDKs.
• Performance Optimizations: Includes built-in acceleration features like CUDA Graph support, static KV caching, and automatic device management to reduce latency.
• Versatile Synthesis: Supports standard text-to-speech, voice cloning from reference audio, and voice design based on descriptive instructions.
• Flexible Deployment: Offers a web UI for manual use, a standalone API mode for integration, and support for streaming audio output to reduce wait times.