Real-time advanced speech enhancement using low-power, fixed-point hardware
Dariusz Bismor, Krzysztof Bartnicki · 2016
This paper describes a project aiming at development of an embedded system speech enhancement solution. It was assumed that a low-cost, low-power, fixed-point processor will be used. It was also assumed that both the stationary and nonstationary noises should be removed. To meet the assumptions, the EZDSP5535 evaluation board, with the TMS320C5535 fixed-point digital signal processor, was selected as a hardware platform. Three speech enhancement algorithms were implemented: the FIR filtering, the adaptive line enhancement, and the spectral subtraction. The last algorithm, being a frequency-domain algorithm, utilizes the FFT co-processor available in the TMS320C5535 chip. All the algorithms can work simultaneously. Examples of good final performance are shown.