Alibaba Cloud kicked off its Apsara Conference with a major upgrade to its Qwen-Audio-3.1 voice model family, rolling out three core series: speech-to-text (ASR), text-to-speech (TTS), and real-time voice interaction (Realtime).

The newly launched audio understanding model, Qwen-Audio-3.1-ASR-Next, is built on next-generation architecture. It can now interpret human emotions, music, ambient sounds, and mechanical noises, while handling tasks like sound description, event localization, audio Q&A, and inference. This allows users to ask questions such as "What is the emotional tone of the person in this recording?" and receive accurate, contextual answers.

On the creation side, Qwen-Audio-3.1-TTS-Next debuts with a fresh architecture that generates complete audio tracks—combining dialogue and environmental sounds—directly from a script. This feature significantly streamlines professional workflows for audiobooks, film and TV productions, and podcast creation.

The real-time interaction capabilities of Qwen-Audio-3.1-Realtime have also been enhanced, enabling the model to listen, think, and respond simultaneously. Improvements include stronger multilingual interaction, role-playing, and empathy features.

These upgraded models are now integrated into Qianwen Office and Qoder, and have been adopted across a range of hardware products, including the QwenNote A2 personal assistant, the QwenNote Eva desktop robot, and Qianwen AI glasses.

Additionally, the simultaneous interpretation model Qwen3.8-LiveTranslate made its first appearance. While human interpreters typically face an average delay of over 4 seconds, this model compresses latency to under 2.5 seconds. It has already been embedded into devices such as Qianwen AI glasses, QwenNote A2, and DingTalk earbuds.

These moves signal Alibaba's accelerating push to bring large language models directly to end-user devices. During the conference, the company also announced Qwen Intelligence, a full-stack AI smartphone solution. Tailored for mobile scenarios, it provides partners with an agent technology platform built on the Qwen models, enabling phones to reliably execute complex cross-app tasks.