An evolutionary algorithm to optimize the microphone array configuration for speech acquisition in vehicles

• Speech enhancement with microphone arrays depends on the array configuration.
• A tailored evolutionary algorithm is proposed to optimize the array configuration.
• A car model is proposed, based on real measurements of background noise.
• Optimized arrays perform much better than common linear arrays.

Speech acquisition using microphone arrays is included in a variety of trending applications. Multichannel speech enhancement based on spatial filtering aims at improving the quality of the acquired speech. The optimization of the filter coefficients has been the primary focus in beamformer design. However, the array configuration plays an important role in the quality of the speech acquisition system and it should also be optimized. In some applications, the possibilities for microphone placement are very large, and the search of the optimum solution, which involves exploring all possible microphone configurations, is an unfeasible task. This work presents a novel search algorithm based on evolutionary computation to approximate the optimum array configuration. A realistic car noise model based on real measurements is proposed and used in the design. The obtained results support the suitability of the method, notably improving the results obtained by linear arrays with the same number of elements, which are the typical arrays currently assembled in vehicles.