Alibaba Cloud has launched CosyVoice, a new AI voice input method that translates dialects into standard Mandarin and can sense what's on your screen.
The company has quietly rolled out the app on macOS and on the app stores of Huawei, Xiaomi, and vivo, where it's currently in a free trial. It takes its name from Alibaba's CosyVoice speech model, while Alibaba's Tongyi Qianwen (Qwen) large language model provides the semantic understanding underneath — enabling voice input across just about every scenario.
The interface is ultra-minimal: just four physical keys — delete, newline, @, and space. If you actually want to type letters or numbers, you have to switch back to the system keyboard.
Thanks to Qwen, voice input automatically cleans up filler words like "um" and "uh," and it can recognize corrections like "no wait, change that to …"