Miniaturized MEMS cluster for speech capture

The last few years has seen a rapid growth in speech recognition markets. Whereas today system works well in quiet homes and office spaces, much better microphone technology for audio focusing is required to make speech recognition robust and useful in complex real-life settings. Broad adoption also dictates that such microphones must be cost effective. Increasing the performance of a microphone array has been traditionally achieved by increasing the size of the array and the number of microphones in the array, resulting in strong trade-offs in performance during product integration. SensiBel proposes to use another approach: using microphones with ultra-high sensitivity in a small, integrated microphone array, which we call a microcluster. This concept is only made possible by sensiBel's unique optical MEMS technology, but will require the microphone elements in microcluster to work very precisely together. This is the challenge that we propose to solve in this project. During the project's first year, we have designed and initiated fabrication of a new batch of miniaturized microphones that was completed in the first quarter of 2019. Further miniaturization was demonstrated in a new batch of microphones based on a new fabrication process and that was delivered in the second part of 2020. These microphones were used to build and characterize progressively smaller arrays. In parallel, we have built several microphone arrays based on earlier samples and smaller samples produced in the project. These have been used to study algorithms for beamforming and calibration. The experimental results confirm the benefit of using high-sensitivity microphones in a small array, but also show the importance of calibration. To achieve a robust solution we have focused on self-adaptive calibration algorithms that can handle changes in the environment. The first version of the algorithms has been implemented and demonstrated in 2019. Further testing and development were carried out in 2020 and 2021. In addition, the self-calibration algorithm has been implemented in an ASIC which achieves very low power consumption

sensiBel has miniaturized its microphone samples to a form factor compatible with very small microphone arrays (<2cmx2cm) The miniaturization has not impacted the SNR. sensiBel has developed a self-calibration algorithm that can operate in the background without intervention from the user. The algorithm has been implemented on an ASIC integrated in the microphone sensiBel has build several small-size microphone-array demonstrators that have been used for industry evaluation. The work has shown the value of high SNR microphones for small arrays Squarehead has evaluated sensiBel's optical microphones for a larger array. The results show that the optical microphones give superior directional performance for lower signal frequencies (due to higher SNR).

The last few years has seen a rapid growth in speech recognition markets. Whereas today system works well in quiet homes and office spaces, much better microphone technology for audio focusing is required to make speech recognition robust and useful in complex real life settings. Broad adoption also dictates that such microphones must be cost effective. Therefore, we wish to develop a new combined hardware and software solution - a miniature microphone array - that meets emerging requirements on performance, size and cost. The aim in this project is to develop a compact, plug-and-play (i.e. “beamforming ready”) microphone array component based on sensiBel's patented optical MEMS microphone technology. The ultra-high SNR of our microphone technology, combined with advanced audio-focusing algorithms, enables very tight spacing of the microphones, making such a single, small array component possible.