Provided by the present invention are a voice processing method, apparatus and terminal device, the method comprising: acquiring voice call information for voice call; acquiring user facial feature information corresponding to the voice call information; and simulating, according to the acquired user facial feature information, backup voice call information of the voice call information when the voice call information is unrecognizable. By means of the present invention, the problem in existing technology wherein an improved solution for voice call has poor voice recognition due to the presence of severe background noise is solved, thereby achieving the effect of improving the voice call quality.