Alibaba Unveils Upgraded Qwen-Audio-3.1 Voice Model Family at Apsara Conference

At the Apsara Conference on September 22, Alibaba announced a comprehensive upgrade to its Qwen-Audio-3.1 voice model family, covering automatic speech recognition (ASR), text-to-speech (TTS), and real-time voice interaction.

The company introduced Qwen-Audio-3.1-ASR-Next, an audio understanding model built on next-generation architecture that can interpret emotions, music, ambient sounds, and mechanical noise, enabling tasks such as sound description, event localization, audio question answering, and reasoning. Also unveiled was Qwen-Audio-3.1-TTS-Next, which uses a new architecture to generate complete audio tracks including dialogue and environmental sounds from a single script, potentially improving production efficiency for audiobooks, films, and podcasts.

Alibaba said the real-time interaction model Qwen-Audio-3.1-Realtime has been enhanced with better multilingual interaction, role-playing, and empathy capabilities. The models have been applied in Qianwen Office and Qoder, and integrated into hardware products including the QwenNote A2, QwenNote Eva desktop robot, and Qianwen AI glasses.

The company also showcased Qwen3.8-LiveTranslate, a simultaneous interpretation model that compresses latency to under 2.5 seconds, compared with an average of over 4 seconds for human interpreters. It has been integrated into Qianwen AI glasses, QwenNote A2, and DingTalk earphones. Additionally, Alibaba said its Qwen Intelligence full-stack AI phone solution would be released during the conference.