Biologically inspired speech processing

点击翻译按钮获取中文摘要

Having successfully developed biologically-inspired multi-microphone speech processing systems in the past four years, we seek to further develop biologically inspired speech localization, enhancement, and recognition systems. Our ultimate goal is to develop automatic speech recognition systems that can understand the spoken words of a human speaker reliably and in practical conditions. While humans can perform such speech recognition with relative ease, mimicking this capability in computer hardware has been an elusive goal.   Today's speech recognition systems are unreliable in the presence of noise and reverberations, making their practical utilization on cell phones, hand held computers, and in cars impossible. Consequently, the fundamental question to ask is: what are the biological processes that allow sound waves arriving at the ears to be correctly recognized, even in the presence of noise? In answering this question, our primary two goals are 1) to study biological sound processing systems (from those of insects to those of humans), and 2) to replicate these processes in electronic hardware. The second goal also requires further study and a careful implementation. In our previous development of a dual-microphone artificial speech processing system, we found that standard computer processors were inadequate in terms of computational power and overall design effectiveness. As a result, we utilized custom-designed hardware implementations as well as reconfigurable hardware to develop new optimized processing structures. Regarding our second goal, a similar implementation study, exploring both custom hardware implementation options as well as reconfigurable hardware, will be conducted. In the end, we seek to develop real-time and reliable speech processing systems that can significantly alter how humans interact and communicate with computers.