Microphone Configuration for Beamformer Design Using the Taguchi Method

In beamformer design, the microphone configurations which represent microphone number and positions are necessary to be optimized in order to improve the effectiveness of speech enhancement. Determination of microphone configuration, number of elements and positions is a nonlinear and non-convex NP-hard optimization problem which was not specified before. However, this is a nonlinear and non-convex NP-hard optimization problem. Gradient-based optimization methods can only converge to suboptimal solutions. Although the recently developed heuristic methods may find better configurations, they require long convergence time. In this paper, we study the effectiveness of using Taguchi method to determine microphone configuration. The Taguchi method is a robust and systematic optimization approach for designing reliable and high-quality models. The method conducts systematic trials based on an orthogonal array which represents a subset of representative configurations. It determines the configurations based on the experimental trials, while the heuristic methods determine the configurations by searching through the configuration domain until no better configuration can be found. A case study based on a common office environment is used as an example to illustrate the effectiveness of the Taguchi method and the commonly used heuristic methods. The numerical results demonstrate that the method is capable to develop the microphone configurations with similar performance compared with the heuristic methods when short computational time is only available. Hence, the method is a strong candidate to design microphone configurations when short development time is only available.