FPGA implementation of spectral subtraction for in-car speech enhancement and recognition

The use of speech recognition in noisy environments requires the use of speech enhancement algorithms in order to improve recognition performance. Deploying these enhancement techniques requires significant engineering to ensure algorithms are realisable in electronic hardware. This paper describes the design decisions and process to port the popular spectral subtraction algorithm to a Virtex-4 field-programmable gate array (FPGA) device. Resource analysis shows the final design uses only 13% of the total available FPGA resources. Waveforms and spectrograms presented support the validity of the proposed FPGA design.