The innovation is to develop very small audio subsystems for mobile audio devices with unprecedented noise and interference reduction. Audio subsystems are fundamental to many high-volume portable devices such as mobile phones, media tablets, and digital voice recorders. Multiple microphones can enhance audio capture and improve signal quality by reducing acoustic noise interference, but current arrays of separated omni-directional microphones require large spacing between microphones to achieve high discrimination. The small size of mobile communication devices thus fundamentally degrades the performance of existing array technologies. "Zero-aperture" mixed-response-type arrays that retain high directional selectivity with very closely spaced microphones coupled with innovative signal-processing methods may achieve unprecedented background noise and interference reduction with high target signal quality. The research will extend existing free-field arrays and adaptive beamforming methods to the different acoustics when embedded in handheld devices, will develop a novel real-time robust beamformer for small arrays, and will greatly increase noise and interference reduction through new multichannel noise-reduction processing. Successful research could fundamentally improve the signal capture and noise rejection of small mobile electronic voice communication devices, thus greatly improving their most fundamental function and enabling their effective use in a wider range of environments.
The broader/commercial impact is to enhance and facilitate mobile voice communication for mobile device users. The voice and audio subsystem is a fundamental component of mobile devices including cell phones, smartphones and media tablets. Background noise interference during voice communication presents a major problem for many users, especially the millions of hearing impaired and elderly individuals who often struggle to communicate using current voice communications technology. Major device manufacturers have recently begun to implement microphone arrays in commercial devices, yet current voice capture systems fail to utilize the full potential of array technology for mobile applications. Market research has predicted that dedicated voice processor sales will grow to 1.6 billion units in 2015 from 63 million in 2010. Prior research has enabled the development of innovative mixed-response-type array technology, allowing arrays to be implemented effectively in smaller devices. However, a number of unsolved technical problems have prevented commercialization of the technology within the mobile market. This Phase I project will address these problems to deliver a disruptive voice processing solution, setting a new standard for voice capture performance in mobile devices.
This Small Business Innovation Research (SBIR) Phase I project developed a very small, microphone-array-based audio subsystem providing unprecedented noise and interference reduction. At about the size of a postage stamp, Sonistic's beamforming microphone system is designed for integration into portable electronic devices such as smartphones, tablets, and notebook computers. The innovation is to improve user experience when voice calling or conferencing in noisy environments, and to enhance speech recognition accuracy by increasing the signal-to-noise ratio. Adaptive beamforming technology coupled with a patent-pending noise-reduction stage developed under NSF funding delivers large-array performance in a package that integrates easily into mobile devices and audio peripherals. This project builds upon more than ten years of academic research in biologically-inspired, acoustic beamforming. Specific focus was given to removing technical impediments which have presently limited transfer of the technology to the commercial sector. The overall technical objective was to develop a new approach delivering a disruptive advance in audio quality via noise and interference reduction, particularly for small mobile devices. Research and development activity focused on producing a commercial demonstration prototype, including compact microphone array hardware and low-latency audio processing software. Prototype hardware is comprised of four standard, low-cost microelectromechanical (MEMS) microphones arranged in a highly compact form factor of approximately 1 in². Software system components include adaptive beamforming for noise cancellation and voice-clarity enhancement, novel multi-channel noise reduction for reducing non-stationary interferers and reverberations, and single-channel noise reduction for further reduction of stationary noise. Real-room testing of the prototype system shows signal-to-noise ratio (SNR) gains in excess of 20 dB over non-stationary interference. Computational analysis and benchmarking indicate that the software, once fully-optimized, will require less than 100 million instructions per second (MIPS), making it ideal for low-power, embedded applications. As a front-end to voice user interfaces (VUI), the system provides advanced automatic speech recognition (ASR) enhancement. Further testing placed the prototype system as an input to Google's standard speech recognition application programming interface (API). Results demonstrate that equivalent word accuracy can be acheived at 15 dB higher noise level using Sonistic's noise suppression as a front-end. Microphone array processing for interference suppression and speech enhancement has been a major topic of research for many decades. Despite significant advancements, background noise still impacts voice communication and interfacing in many of today's most advanced products. Traditional large-aperture microphone arrays are ill-suited for integration into portable, mobile, and wearable electronics. This research has demonstrated for the first time the ability to greatly suppress noise and nonstationary interference using a small microphone array of less than an inch square, in real-world environments, with reverberation and a variety of natural sources. Voice communication remains a primary mode of communication for consumers, and voice interfaces and control of computers and computer-based systems is rapidly increasing in importance. For example, automatic speech recognition is predicted by Frost and Sullivan to be a standard feature in cars by 2020. Better hands-free communication and voice-activated control in automobiles could lower driver distraction, reducing the number of automobile accidents and deaths. Reliable hands-free voice-activated communications in noisy environments could improve the effectiveness and safety of public-safety personnel such as firemen and policemen, as well as soldiers on the battlefield. Successful commercialization of this technology has the potential to enhance voice capture in mobile devices, facilitating communication for billions of users worldwide.
