Evaluating the performance of a multimodal speaker tracking system at the edge-to-cloud continuum