v0.1.5

Changes

• All Linux builds now use libtorch from libtorch-releases and bundle it in the release zip
• Added ARM64 + CUDA (Jetson) build
• install.sh and bootstrap.sh detect CUDA on both x86 and ARM
• No separate libtorch download needed at install time

Assets

Asset	Platform
qwen3-tts-macos-aarch64.zip	macOS Apple Silicon (MLX)
qwen3-tts-linux-x86_64.zip	Linux x86_64 CPU
qwen3-tts-linux-x86_64-cuda.zip	Linux x86_64 CUDA
qwen3-tts-linux-aarch64.zip	Linux ARM64 CPU
qwen3-tts-linux-aarch64-cuda.zip	NEW Linux ARM64 CUDA (Jetson)

v0.1.4

Add install.sh installer, Linux x86 CUDA release build, and pre-generated tokenizers for all models

v0.1.3

What's Changed

• Fix MLX softmax precision for stable TTS sampling by @juntao in #12

Full Changelog: v0.1.2...v0.1.3

v0.1.2

Fix

• Include mlx.metallib in the macOS ARM64 release tarball so the binary works on user machines without the CI runner's build path

v0.1.0

Qwen3 TTS Rust v0.1.0

First release of the Rust port of Qwen3 TTS text-to-speech.

Features

• Text-to-speech with Qwen3-TTS-12Hz-0.6B-CustomVoice and 1.7B-CustomVoice models
• Voice cloning with Qwen3-TTS-12Hz-0.6B-Base model (standard and ICL modes)
• Dual backend support: LibTorch (CPU/CUDA) and Apple MLX (Metal GPU)
• Built-in speaker presets (Vivian, Ryan, etc.)
• English and Chinese language support
• Emotion/style instruction support (1.7B model)

Downloads

• Linux x86_64 CPU: qwen3-tts-linux-x86_64-cpu.tar.gz
• Linux x86_64 CUDA: qwen3-tts-linux-x86_64-cuda.tar.gz
• Linux ARM64 CPU: qwen3-tts-linux-aarch64-cpu.tar.gz
• macOS ARM64: qwen3-tts-macos-arm64.tar.gz (MLX backend, Apple Silicon)

Binaries require libtorch to be installed separately on Linux. See README for setup instructions.

v0.1.1

Fix

• Include mlx.metallib in the macOS ARM64 release tarball so the binary works on user machines without the CI runner's build path