Voice Quality Improvement with Error Concealment in Audio Sensor Networks

摘要

Multi-dimensional properties of audio data and resource-poor nodes make voice processing and transmission a challenging task for Wireless Sensor Networks (WSN). This study analyzes voice quality distortions caused by packet losses occurring over a multi-hop WSN testbed: A comprehensive analysis of transmitted voice quality is given in a real setup. In the experiments, recorded signals are partitioned into data segments and delivered efficiently at the source. Throughout the network, two reconstruction scenarios are considered for the lost segments: In the first one, a raw projection is applied on voice with no error concealment (V-NC) whereas the latter encodes a simple error concealment (V-EC). It is shown that with an affordable reconstruction, a comprehensible voice can be gathered even when packet error rate is as high as 30%.