Artem Voloskovets
⬤ Alibaba's Tongyi Lab just dropped its Qwen3-TTS lineup, and it's packing two interesting speech tools: VoiceDesign and VoiceClone. These additions beef up the Qwen ecosystem with voice customization and quick cloning features that work across multiple languages. The whole thing's aimed at developers and businesses building the next wave of voice-powered AI and interactive systems.
⬤ VoiceDesign (or VD-Flash, if you prefer) lets you whip up custom voices just by describing what you want in text—think tone, rhythm, personality, the works. No need to pick from a limited menu of preset voices anymore. Tongyi Lab says VoiceDesign crushes it on role-play benchmarks, beating out both GPT-4o-mini and Gemini-2.5-pro in head-to-head tests. It's built for situations where you need voices that actually express something, whether that's role-playing scenarios or chatbots that don't sound like robots.
⬤ Then there's VoiceClone, branded as VC-Flash, which does exactly what it sounds like—instant voice copying. The wild part? Tongyi Lab claims it only needs three seconds of audio to clone someone's voice. As Tongyi Lab puts it, the system "can clone a voice from as little as three seconds of audio input" while handling ten different languages. They're also reporting a 15 percent lower error rate compared to ElevenLabs when testing across multiple languages, which suggests they're serious about keeping quality consistent no matter where you're using it.
⬤ This release shows how heated the competition's getting in AI voice tech, with everyone racing to nail controllability, accuracy, and multilingual capabilities. For Alibaba and the market at large, tools like VoiceDesign and VoiceClone signal that speech generation isn't just a feature anymore—it's becoming essential infrastructure. As Alibaba keeps expanding its Qwen family, these developments could reshape how digital media, customer service platforms, and AI content creation handle voice, setting new standards for what scalable and customizable voice tech should look like.
Artem Voloskovets
