On May 28, 2026, the authoritative AI evaluation platform Artificial Analysis released the latest Speech Arena rankings. Alibaba made a strong breakthrough with its voice large model Fun-Realtime-TTS-Preview, securing the fifth place globally and the top spot domestically with an Elo score of 1190.

1. Comprehensive Leadership: Dominating Three Core Voice Tracks

In this evaluation, Alibaba's voice technology system demonstrated strong overall capabilities, ranking first in the country across three key voice AI tracks:

  • ASR (Automatic Speech Recognition): It ranked first nationally in accuracy and robustness in converting speech to text, representing Alibaba's understanding ability in complex audio environments.

  • Chat (End-to-End Speech Understanding and Dialogue): It won the title of first in the national ranking for real-time voice conversation fluency, logic, and response speed, marking Alibaba's top industry level in intelligent assistant interaction.

  • TTS (Text-to-Speech): As a core competitive area, Fun-Realtime-TTS-Preview not only set new domestic records in naturalness, emotional expression, and rendering speed, but also established a benchmark on a global scale.

2. Technological Breakthrough: The Real-Time Advancement of Fun-Realtime

The key player in this ranking—Fun-Realtime-TTS-Preview—represents a major breakthrough by Alibaba's voice team in real-time speech synthesis.

Previously, speech synthesis often faced the dilemma of being unable to achieve both high naturalness and fast response. However, Alibaba's model successfully achieved speech output with near-human intonation under millisecond-level latency through an end-to-end deep architecture. This real-time capability holds decisive significance for scenarios requiring high timeliness such as smart car interactions, digital human live streaming, real-time translation, and customer service.

3. Industry Insights: Domestic Voice Technology Moving Toward "Deep Intelligence"

As a trendsetter in the AI field, Artificial Analysis has an extremely strict scoring system, not only testing model performance on test sets, but also emphasizing user experience in real-world scenarios. Alibaba's "three championships" are not just about scores, but also convey the following core messages:

  1. Speech AI Enters the "Large Model Era": Previous speech technologies mainly relied on traditional statistical methods or small models, while Alibaba's success proves that integrating speech processing into a deep learning large model foundation can bring a significant leap in perception quality.

  2. "Chinese Speed" in Scenario Implementation: With Alibaba leading in both speech understanding and generation, future domestic smart hardware and large model ecosystems will have stronger global competitiveness in the core entry point of "voice interaction."

  3. Illustration of Closed-Loop Capabilities: From recognition (ASR) to understanding (Chat) and then to synthesis (TTS), Alibaba has completed the full voice interaction pipeline, laying a solid infrastructure for building seamless AI intelligent agents (Agents).

With continuous bottom-up technical layout and model iteration in the voice field, domestic AI is accelerating toward deeper waters—from "being able to recognize" to "understanding human emotions and interaction logic" more deeply.