Alibaba made a dual push into the AI voice market on August 7. Its enterprise collaboration platform, DingTalk, launched a full-scenario AI voice input feature, becoming the first application to achieve comprehensive coverage within a collaboration platform. Simultaneously, Alibaba introduced CosyVoice Studio, China's first all-in-one AI voice productivity platform, featuring three core modules: voice recording, AI agents, and audio creation. Both products are built on Alibaba's proprietary Qwen-Audio model, which swept the top spots globally in speech recognition, real-time interaction, and text-to-speech on the Artificial Analysis benchmark, achieving a character error rate of just 1.7% and outperforming international rivals like GPT-Realtime-2. DingTalk focuses on penetrating consumer office scenarios, while CosyVoice Studio targets the enterprise market and content creation, adopting a tiered rollout strategy with some features available via a whitelist invitation system. This marks the first time Alibaba's voice model capabilities have been offered externally in an aggregated product format, establishing a dual-engine AI voice strategy driven by both consumer and business segments.

Key Elements

Alibaba played two critical cards in the AI voice arena simultaneously. On August 7, the group's enterprise collaboration platform, DingTalk, announced the launch of an intelligent voice input feature, becoming the industry's first collaboration application to support full-scenario AI voice input. On the same day, Alibaba officially launched CosyVoice Studio, China's first all-in-one AI voice productivity platform, making its proprietary voice model capabilities available externally in an aggregated product format for the first time.

Both launches are built on the technical foundation of Alibaba's proprietary large language model, Qwen-Audio. According to a report by Time Weekly, DingTalk's newly launched intelligent voice input feature integrates Alibaba's proprietary speech recognition model, Qwen-Audio-3.0-ASR-Flash-Message. On Artificial Analysis, a globally authoritative AI evaluation platform, the Qwen-Audio model family ranks first globally with a character error rate of just 1.7%, surpassing top international models like GPT-Realtime-2. More notably, the model simultaneously secured the number one global ranking across three categories on Artificial Analysis: Automatic Speech Recognition (ASR), Real-Time Interaction, and Text-to-Speech (TTS).

In terms of functional coverage, DingTalk's voice input upgrade has penetrated every core scenario on the platform. The feature now fully covers high-frequency entry points within DingTalk, including Tongyi Qianwen Office, instant messaging, the search bar, documents, and AI spreadsheets. More groundbreaking is that this capability also supports use in any environment requiring text input, such as external browsers and desktop AI agent platforms. This means users can invoke its speech-to-text service even outside DingTalk's closed ecosystem.

CosyVoice Studio, launched the same day, pushes Alibaba's voice AI capabilities into a broader range of commercial scenarios. The platform includes three core functional modules: CosyFlow for voice recording, CosyAgent for voice AI agents, and CosyCreative for audio content creation tools.

CosyFlow layers semantic understanding and text generation capabilities on top of traditional speech-to-text conversion. In daily scenarios, the feature can automatically filter out filler words like "um" and "uh," organizing fragmented thoughts into clear expressions. In work scenarios, users can dictate to generate formatted emails, work reports, and meeting minutes. The built-in "Quick Notes" feature is designed for meetings, interviews, and lectures, supporting real-time recording and offline uploads. It can distinguish between different speakers based on voiceprints and automatically generates structured chapters and intelligent summaries after recording ends.

CosyAgent allows users to quickly create AI agents equipped with enterprise knowledge, tool-calling capabilities, and real-time voice interaction through natural language. It is compatible with prompt and workflow configurations and is primarily aimed at enterprise-level scenarios such as intelligent customer service and telemarketing. CosyCreative features a built-in library of thousands of voice timbres, supports voice cloning, and generates highly realistic voices. Users can upload documents or web links, or input text to generate conversational podcasts or multi-character audiobooks, covering the entire audio content creation chain.

It is worth noting the distinct rollout strategies for the two products. DingTalk's intelligent voice input feature is directly integrated and launched for its massive existing user base with an extremely low barrier to entry. In contrast, CosyVoice Studio adopts a tiered open strategy—users can search for "CosyVoice" in app stores starting today to download and experience it for free for a limited time. However, CosyAgent and CosyCreative are currently being tested by enterprise users via a whitelist invitation system and will be fully opened to the public soon.

From a technology-to-product pathway perspective, this marks the first time Alibaba's voice model capabilities have been made available externally in an aggregated product format. Enterprise users can directly experience the results on the platform and then call APIs on demand, while individual users can also use the product directly. This "experience first, integrate later" model helps lower the technology evaluation costs for enterprises and accelerates the commercial deployment of AI voice capabilities.

The simultaneous launch of these two major products demonstrates that Alibaba is accelerating the transformation of its large AI model capabilities from technical reserves into scenario-based applications accessible to end-users. DingTalk, serving as a collaboration entry point with hundreds of millions of users, is tasked with penetrating AI voice capabilities into daily office scenarios. Meanwhile, CosyVoice Studio plays the role of exporting capabilities to the enterprise market and the content creation sector. The two complement each other, jointly building Alibaba's dual-engine "Consumer + Business" strategy in the AI voice field.