Alibaba has released Qwen-Audio-3.0-TTS, a new AI dubbing model that can control emotion, speed, and character on command — complete with laughs, gasps, coughs, and sighs.
Users just input something like “say it angrily” or “laugh halfway through,” and the model follows direction. Where past AI dubbing was mostly about reading text cleanly, this one actually understands directorial notes.
The Plus version supports 16 languages and 20 Chinese dialects, generating audio up to 48kHz. Even if the reference recording has background noise, it can extract the voice’s timbre and perform voice cloning.
In a third-party blind test by Artificial Analysis, the Plus version ranked first.