Abstract: Techniques are disclosed for selecting one or more microphones on a remote speaker/microphone (RSM) assembly for input to an audio processing algorithm. An up orientation for the RSM is determined based on an embedded orientation sensor. For each microphone on the RSM, an up distance relative to the orientation sensor is determined based on a stored fixed mapping of microphone locations and current orientation of the orientation sensor, the microphone locations being fixed and known relative to the orientation sensor. Each microphone on the RSM is ranked from the uppermost to the lowermost. The uppermost microphone on the RSM assembly is then selected to serve as a voice input microphone while the microphone located furthest from the voice input microphone is selected to serve as a noise input microphone.