Deep Neural Network based Distance Estimation for Geometry Calibration in Acoustic Sensor Networks

Abstract

We present an approach to deep neural network based (DNN-based) distance estimation in reverberant rooms for supporting geometry calibration tasks in wireless acoustic sensor networks. Signal diffuseness information from acoustic signals is aggregated via the coherent-to-diffuse power ratio to obtain a distance-related feature, which is mapped to a sourceto-microphone distance estimate by means of a DNN. This information is then combined with direction-of-arrival estimates from compact microphone arrays to infer the geometry of the sensor network. Unlike many other approaches to geometry calibration, the proposed scheme does only require that the sampling clocks of the sensor nodes are roughly synchronized. In simulations we show that the proposed DNN-based distance estimator generalizes to unseen acoustic environments and that precise estimates of the sensor node positions are obtained.

Iterative Geometry Calibration from Distance Estimates for Wireless Acoustic Sensor Networks

“…The acoustic sources are simulated using speech signals from the TIMIT database [18], which are convolved with the RIRs to obtain the microphone signals. The DNN approach from [8] is used to estimate the distance between sensor node and acoustic source.…”

“…Following the ideas of [10], the localization procedure has to select one sensor node as reference node, which has to be chosen judiciously, as the resulting localization error depends on the precision of the distance estimate of the selected sensor. Here, we always select the node with the smallest distance estimate to the acoustic source as reference node, as it was shown in [8] that the distance estimation error increases with the distance between audio source and sensor. Informal experiments showed an overall error reduction by approximately a factor of two compared to randomly selecting the reference node.…”

“…The DNN-based distance estimates are corrupted by an error which tends to be heteroscedastic (see [8]). So WLS is employed to estimate the location Ôk with…”

“…In recent publications [6][7][8] it was shown how an estimate of the distance between the acoustic source and the microphone array can be obtained from the recorded audio signal. Distances as inputs allow to employ techniques for geometry calibration which were previously used to localize mobile devices based on signal strength information.…”

“…In this paper we present a geometry calibration algorithm that uses source-node distance estimates from a deep neural network (DNN) based acoustic distance estimator [8]. The distance estimation method does not required the nodes of the sensor network to be time synchronized, which is a major advantage over many other geometry calibration approaches.…”