The present disclosure discloses a voice interaction method, system, terminal device and medium. The method comprises the following operations performed by a terminal device: performing voice recognition on collected voice signals to acquire an input sentence; semantically matching the input sentence with cached sample sentences to determine whether there is a sample sentence having the same or similar semantics as the input sentence among the cached sample sentences; if yes, acquiring cached response content corresponding to the sample sentence having the same or similar semantics as the input sentence as response content of the input sentence; if not, sending the input sentence to a server; receiving from the server response content acquired by the server through semantic understanding according to a knowledge base stored on the server; and responding to the input sentence according to the response content.