DeepSeek is reportedly testing limited voice playback in its app: some users can hear responses read aloud and choose from four voice profiles. The feature remains a gray test rather than a full public launch, and it has not been established as a hands-free, two-way voice-chat mode.
The limited test was first reported on September 12, 2026.
What some test users reportedly see
The reported interface includes a small speaker button in the upper-right corner of the DeepSeek app. Tapping it lets some test users play a spoken version of a response.
The app settings also include four selectable profiles. Their reported Chinese labels are 贝壳, 白浪, 海星 and 暗潮, with descriptions broadly corresponding to lively and changeable, bright and firm, playful and sweet, and deep and resonant voices. They are profiles within the test—not four separate products and not voices confirmed for every user.
So, can DeepSeek do voice chat? The practical answer is narrower: some users can reportedly listen to DeepSeek responses. Hearing an answer aloud does not by itself amount to a continuous conversation in which the user speaks naturally and receives spoken replies. DeepSeek’s test has not been announced as a public voice mode.
External developers can build a separate voice interface
Developers can connect DeepSeek’s text API to two additional components: speech-to-text, which turns spoken words into text, and text-to-speech, which turns the model’s written answer back into audio. That arrangement can create a voice interface around DeepSeek, but it is a developer-built system rather than proof that the native app offers the same capability.
This distinction matters because a tutorial or integration built with external voice services may include features such as spoken input, wake-word detection or continuous exchanges. Those functions belong to that external pipeline. They should not be read as features of the limited native DeepSeek app test.
For now, the meaningful change is modest but clear: DeepSeek is experimenting with letting some app users listen to answers and pick among four voices. It is a step toward richer interaction—not a confirmed public, two-way voice assistant.
