The present disclosure provides a speech processing method. A specific implementation solution is: a terminal device sends at least one speech intention to a server in a process of receiving first speech information, where each speech intention is a speech intention corresponding to a part of speech information in the first speech information; the server acquires response information corresponding to the at least one speech intention; the terminal device sends the first speech information to the server in response to completion of receiving the first speech information; the server acquires a second speech intention corresponding to the first speech information, and sends the response information corresponding to the first speech intention to the terminal device, and the terminal device outputs the response information.