FFT-based daul-ALU processor for speech enhancement

Speech enhancement is very important in the application of speech recognition and communication. To implement it, a FFT-based dual-ALU processor by using hardware-software codesign is proposed in this paper. For the consideration of computation load, the speech enhancement adopts spectral subtraction method to reduce the influence of noise. In this case, FFT occupies most of the computation load. To reduce the computation load and power consumption, we proposed radix-2 pipeline FFT architecture. Based on this FFT, the spectrum analysis can be efficiently applied. Then we proposed a dual-ALU processor to implement the spectral subtraction. This dual-ALU architecture can provide parallel calculation capability. Besides, the area of the second ALU is only half of the first ALU. The design has been fabricated in the total area of 2.82 mm times 2.82 mm by using TSMC 0.13 mum 1P8M CMOS technology.