A method for speech switching, including: extracting mute flags from encoded speech data transmitted by each of the terminals respectively in order to determine one or more non-mute terminals; decoding the encoded speech data of each non-mute terminal respectively; calculating speech energy of each non-mute terminal according to the decoded speech data of each non-mute terminal; comparing the speech energy of the non-mute terminals and selecting one or more terminals with relatively large speech energy; performing linear superposition with different combination methods to the decoded speech data of the selected terminals, and encoding the decoded speech data which is obtained from the linear superposition of the different combination methods, and transmitting the data to the corresponding terminals respectively. An apparatus for speech switching is also disclosed. With this invention, resource consumption during procedure of speech switching may be reduced, the effect of speech communication may be improved.