QwenVoice brings offline Qwen3-TTS to macOS

Wrapping MLX inference in a native SwiftUI app is a smart pattern for bringing complex open-weights models to end users without friction.

• –Bundling the Python runtime inside the DMG is a crucial step for local AI adoption, removing dependency headaches for non-developers.
• –Using the 8-bit quantized 1.7B model makes advanced TTS accessible even on entry-level 8GB Apple Silicon machines.
• –The two-process architecture (SwiftUI frontend communicating with a Python backend via JSON-RPC) provides a robust blueprint for future local-first AI applications.
• –On-device voice cloning and generation capabilities offer a compelling, privacy-first alternative to paid cloud TTS providers.