The embodiment of the invention provides a voice endpoint detection method and device. The method comprises the steps of: performing framing processing of voice to be detected to obtain a plurality ofvoice frames to be detected; obtaining acoustic features of each voice frame to be detected, and inputting the acoustic features of each voice frame to be detected into a voice activity detection (VAD) model, wherein the VAD model is configured to classify previous N voice frames in the voice to be detected into noise frames, classify the (N+1)th voice frame to the last one voice frame to voice frames and classify the M noise frames after the last voice frame to the voice frames, and N and M are integers; and determining a starting point and a destination of the voice section according to theclassification result output by the VAD model, wherein the starting point corresponds to the voice frame firstly classified as the voice frame, and the destination is a voice frame classified to thevoice frame at the last. The accuracy of voice endpoint detection can be improved.