On the Integration of Sampling Rate Synchronization and Acoustic Beamforming

Time synchronization of the nodes of an acoustic sensor network is important to be able to apply acoustic beam-forming for signal extraction. While former solutions perform the tasks of synchronization and beamformer coefficient estimation in cascade, we here show how they can be carried out jointly. The key observation is that the spatial covariance matrices of the speakers contain all information necessary to carry out both tasks. We simulate a meeting transcription system with asynchronous sensor nodes and show that the joint treatment not only saves some computations, but also leads to improved sampling rate offset estimation and beamforming. However, the final transcription performance turns out to be insensitive to those improvements.