Alibaba launches Qwen-Audio-3.0-Realtime, a voice model with dynamic tone and voice cloning

Alibaba has launched the Qwen-Audio-3.0-Realtime voice model, with a Plus version for stronger reasoning and a Flash version for faster speed. It's designed for smart customer service, education, and emotional companionship. The model upgrades in intelligence, tool calling, empathy, and smooth duplex interaction, achieving top results on the S2S voice benchmark VStyle.

Key features: it adjusts tone and emotion based on context, supports voice cloning, and avoids robotic speech. With built-in multimodal duplex control, it focuses on your voice even in noisy places. It ranks first in Artificial Analysis sub-tests, beating OpenAI GPT-Realtime-2.

It also supports dynamic tool calling for tasks like route planning and info searches. In role-play, it adapts speaking style to the character; in emotional chats, it shows empathy through tone and rhythm. Voice cloning is available for custom needs.