Digital signal processing for Smart Microphones

This whitepaper introduces the concept of smart microphone as an evolution of the classical microphone driven by the emergence of natural speech interfaces. From pure sound pressure to electrical signal converters, voice acquisition devices have evolved into complex systems featuring multiple capture channels coupled with digital signal processing capabilities. In this whitepaper, we focus on beamforming noise cancelling smart microphone applications.

Following a reminder of the fundamentals of array processing and adaptive filtering theory, a review of commonly used algorithms for beamforming, voice activity detection and noise/echo cancellation is provided. These algorithms can be combined to form complete solutions for beamforming noise calling smart microphone for a wide variety of applications. Possible implementations on the XMOS xCORE-200 architecture are subsequently analyzed in terms of computational complexity, memory usage and cores distribution. Our investigations show that these solutions based on the well-known HiRes Delay and Sum example should require only modest overhead, even if run at sampling rates as high as 48kHz.  This demonstrates the practical usability of the xCORE-200 architecture for smart microphone applications.