Alibaba Cloud has quietly launched a new AI voice input app called CosyVoice on macOS and through app stores for Huawei, Xiaomi, and vivo phones. It’s currently free to try, and it promises to handle just about any voice input scenario. The name comes from Alibaba’s larger CosyVoice speech model, and underneath, it’s powered by the company’s Tongyi Qianwen LLM for understanding what you actually mean.
The keyboard interface is almost painfully minimal — just four physical keys: delete, newline, @, and space. If you actually want to type, you’ll need to switch back to your system keyboard. But the real smarts are in the voice engine. Tongyi Qianwen automatically cleans up filler words like “um” and “ah,” and even understands spoken corrections like “no, change that to…” Beyond that, it can automatically format parallel items into numbered lists or tables, and it’ll convert Shanghainese, Cantonese, and Sichuan dialect in real time into standard Mandarin. There’s also a hotword feature for importing specialized vocabulary.
There are differences between the two clients. macOS gets a unique feature called Screen Context Enhancement. When you’re dictating, the system will automatically take screenshots and upload them to better understand what you’re talking about, which obviously raises privacy questions. Compared to more established options like WeChat Input or Doubao (ByteDance’s assistant), CosyVoice is still in early iteration. It’s less integrated and less stable, and it has a tendency to over-clean your speech, removing too many natural pauses and repetitions.