Conference mixing using turbo-VAD

A conference mixer includes a unit configured to receive a plurality of input streams, a spectral voice activity detection (VAD) unit configured to, for each of the input streams, generate and output a spectral VAD decision indicating whether a frame including data packets is voice, a turbo VAD unit configured to generate and output a turbo VAD decision that indicates for a frame including data packets which input stream is active, the turbo VAD decision being based on the spectral VAD decisions and a power-based decision indicating whether an estimated instantaneous power level of a frame including data packets is greater than a power threshold, and a finite state machine (FSM) unit configured to select which of the input streams to output as an active stream based on a plurality of the turbo VAD decisions, the turbo VAD decision being based in part on feedback provided by the FSM.