Comparison of two open-loop adaptive speech enhancement algorithms for digital hearing aids

Abstract

For the hearing impaired, listening in noise is one of the biggest problems. With the advent of true digital hearing aids, new solutions for this problem are being researched. This paper presents a comparison of two open-loop adaptive speech enhancement algorithms for application in digital hearing aids based on Texas Instruments digital signal processing (DSP) chips. The first algorithm uses second order signal statistics to update the coefficients of an adaptive Finite Impulse Response (FIR) filter. The second algorithm uses higher order (up to fourth) statistics to update the coefficients. In theory, the adaptive FIR based on higher order statistics should perform better in eliminating Gaussian noise, including pink noise. However the results presented in this paper show that this algorithm's performance degrades rapidly as the input signal-to-noise (SNR) drops below 0 dB. Additionally, if there are multiple frequencies, with different power levels, present in the input, the higher order based adaptive filter greatly distorts the power relationship between the frequencies. In an application like speech, where the relative power levels of the formant frequencies can be critical, this would not be acceptable. Hence, in the final analysis, it is shown that the second order statistics based adaptive filter is preferred for speech enhancement.