Evidence hub
Qwen3 TTS Benchmark: Voice Quality & Cloning Tests
A transparent home for playable examples, test context, source references, and the limits of the evidence currently available on qwen3tts.net.
Test overview
This page is not a numerical ranking or a claim of universal quality. It groups the evidence that is currently publishable: real playable examples, the visible prompt context, and the methodology used to interpret them.
Test environment
Samples linked here were made available through qwen3tts.net’s AI voice workflow. The page does not infer a fixed model version, universal latency, or a configuration that is not visibly documented alongside a sample.
Natural speech samples
These existing examples provide listening context for their specific scripts. They are not scored, and they should not be used to predict every voice, language, or production setup.
Narration & storytelling
Input excerpt: The old lighthouse stood silent against the crashing waves, its beam cutting through the thick fog like a sword.
Corporate & news
Input excerpt: Market trends indicate a significant shift towards renewable energy investments in the coming quarter.
Voice cloning & style control
Official model capabilities are documented in the primary sources below. This benchmark does not add a similarity score or a style-control claim without a published input and corresponding evidence.
Read the voice cloning guideKnown limits
Results can vary with the script, selected voice, language, reference audio, and current implementation. We do not publish fabricated scores, sample counts, or conclusions for configurations that are not evidenced here.
Sources & evidence
Primary sources
- Qwen3-TTS official model repository
Official model documentation · QwenLM
- Qwen3-TTS CustomVoice model card
Official model page · Qwen