Abstract: An electronic timepiece with voice memory includes an auto start circuit which senses the presence of an input voice and automatically controls the start of recording. Voices inputted from the outside are coded and the coded data is stored in memory. An output means reads out the memory contents and transforms the coded contents into an analog voice signal. When the incoming voice exceeds a predetermined level, the recording means is automatically started by an output from a level detector. Memory is divided into at least two regions. The first region stores coded voice data before the auto-start circuit operates and later coded voice data is recorded in the second region after the autostart circuit operates so that the initial portions of speech are not lost.