Bio-inspired, Nonlinear and Adaptive Acoustic Sensing – Study of Sensor Design (vor Ort)
* Presenting author
Abstract:
Bio-inspired signal pre-processing introduced into speech processing systems can increase the performance of these systems regarding e.g. recognition rate remarkably. Thereby, pre-processing mainly refers to frequency decomposition and nonlinear amplification of the signal. In contrast to the inner ear, this pre-processing is applied only after the transduction stage and realized mainly in software. A hardware-based implementation of the functionality directly at the sensor level cannot only enable real-time performance, reduce data streaming to the processing stage, and increase efficiency of the overall system but also improve signal-to-noise ratio and sensitivity particularly in noisy or hard-to-hear conditions. We developed such an acoustic sensor system with integrated bio-inspired signal processing using active silicon beams with integrated deflection sensing and actuation in combination with real-time feedback. While resonant operation enables the frequency decomposition, the feedback enables a wide tuning of sensor properties including switching from linear to nonlinear transfer characteristics. Here, we present our experimental and numerical study of the linear and nonlinear sound response for two different sensor designs.