Alibaba's Qwen-Audio-3.1-TTS Tops Global Voice Leaderboard on Artificial Analysis

On September 28, Alibaba's recently released speech synthesis model Qwen-Audio-3.1-TTS ranked first globally on the Controlled Voice Arena leaderboard of Artificial Analysis, a leading AI evaluation platform, with an Elo score of 1177.

Qwen-Audio-3.1-TTS is Alibaba's next-generation text-to-speech model that supports multilingual and dialect synthesis. It enables a single timbre to transfer naturally across languages, allowing one voice to speak multiple languages. Unlike traditional speech synthesis that mainly focuses on pronunciation accuracy, the model emphasizes adapting the voice to the specific content and usage scenario. Users can control emotion, speaking rate, and delivery style through instructions, making the generated speech sound natural and expressive.

The model is part of the Qwen-Audio-3.1 series unveiled at Alibaba's 2026 Apsara Conference. The series also includes speech recognition (ASR) and real-time speech interaction models, with API services for all three models available on the Qwen AI platform. In addition, several open-source models from Alibaba's audio team, such as Qwen-Audio-Agent and CosyVoice, have accumulated tens of thousands of stars on GitHub and attracted attention from developers worldwide.

Qwen-Audio models have seen price cuts of up to 95%, significantly reducing costs for developers and enterprises in speech-related applications.