实时的Gammatone听感知滤波器组的FPGA实现  

Real-time Implementation of Auditory Gammatone Filter Bank with FPGA

作　　者：贾瑞[1] 李冬梅[2] 

机构地区：[1]清华大学微电子学研究所,北京100084 [2]清华大学电子工程系,北京100084

出　　处：《微电子学与计算机》2015年第1期35-39,共5页Microelectronics & Computer

基　　金：国家高技术研究发展计划(2012AA012301);国家自然科学基金项目(61171001)

摘　　要：Gammatone听感知滤波器组能大幅度的提高数字助听器、语音增强和语音识别等语音处理系统的性能,但因其庞大复杂的计算量以及不可实时的综合方式限制了其实际应用.设计了一种高效的运算单元结构和复用架构,提出了一种通过延时补偿的方法简化其综合的方式,并在FPGA上实现了128通道的便于综合的可实时处理语音信号的Gammatone滤波器组.通过测试验证,该设计延时为20ms,满足实时性要求,能很好地重构语音信号,并提高语音增强系统的性能.Gammatone Filter Bank(GTFB)is perfectly suitable for speech processing systems like digital hearing aids,speech enhancement and speech recognition system.However,the high computational complexity and nonreal-time characteristic in the hardware synthesis restrict its practical application.In this paper,an efficient core computation unit structure is proposed.Besides,a low complexity synthesis algorithm is proposed and the real-time characteristic is achieved by compensating the filter delay.The measurement results verified show that a 128-band high-efficient real-time reconfigurable Gammatone Filter Bank is realized in FPGA.The design has only 20 ms delay and improve the dual-microphone CASA speech enhancement system performance.

关 键 词：Gammatone滤波器组 语音增强系统 延时补偿 实时性 FPGA 

分 类 号：TN912.3[电子电信—通信与信息系统]