Qwen3-TTS

1.7B text-to-speech model by Alibaba covering ten major languages plus dialect voices, built on the 12 Hz Qwen3-TTS tokenizer. Fifteen cards carry a recipe here from a 12 GB floor, and every one of them is set up for multilingual voice cloning rather than plain read-aloud synthesis. Both Radeon boards are covered under ROCm. Apache-2.0.