A method of communicating a speech signal over a communication medium which utilizes VOX (Voice Operated Transmitter) control, and a speech signal transmitting unit and a speech signal receiving unit used in said method. The speech signal includes at least one speech-active duration in which actual speech exists and one silent duration in which no actual speech exists. At a transmitting side, the speech signal is analyzed for each of successive frames thereof to determine whether each frame is in the speech-active duration, and is coded for each of successive frames thereof to produce a series of successive coded data of the speech signal. Then, in response to the detection of a speech starting frame of the speech-active duration, the coded data of the speech starting frame is switched to a preamble, and the preamble and the successive coded data subsequent to the speech starting frame are transmitted in the form of a series of successive frames. At a receiving side, the preamble is detected and the decoded results of the coded actual speech data subsequent to the detected preamble are output as a speech signal. When a predetermined number of successive coded actual speech data have been detected without having detected a preamble preceding thereto, it is decided that the decoded results of the coded actual speech data are in a speech-active duration and the decoded results are outputted as a speech signal.