Abstract: A central dictation system includes a central server computer and a plurality of voice input stations connected to the server computer. Speaker-dependent speech recognition capabilities are provided at each voice input station, and any authorized user may use any station. When a user logs on to one of the stations, acoustic reference files to be used in recognizing the speech of the user are downloaded from the server to the station onto which the user has logged on. The down-loaded acoustic reference files are used to perform the speech recognition at the voice input station. A high bandwidth signal generated at the voice input station and used for speech recognition processing is transcoded at the input station to form a low bandwidth speech signal. The low bandwidth speech signal and the text document resulting from the speech recognition process are uploaded to the server computer.