Alibaba Upgrades New Generation Speech Model Qwen3-TTS, Can Generate Anthropomorphic Tones Based on Text and Sound

This article reports on Alibaba's upgrade to its Qwen3-TTS speech model, introducing VoiceDesign (VD) and VoiceClone (VC) models. The claim that it significantly surpasses GPT-4o in generation effects is noteworthy and requires further validation. The ability to DIY sound design and pixel-level timbre imitation, including enabling animals to "natively" speak human language, suggests significant advancements in speech synthesis. The potential applications in audiobooks, AI comics, and film dubbing are highlighted, indicating a focus on professional applications. The article emphasizes the naturalness, stability, and efficiency of the generated speech, which are crucial factors for real-world adoption. However, the article lacks technical details about the model's architecture and training data, making it difficult to assess the true extent of the improvements.