The frequency domain noise reduction techniques for digital hearing aids

Abstract

The overall performance of digital hearing aids can be degraded by some disturbing sources such as acoustic feedback signal, interfering signal, and background noise, etc. This dissertation focuses on the Noise Reduction (NR) techniques, especially for additive background noise in digital hearing aids. A proposed NR technique demonstrates an improved scheme based on the hybrid structure between two existing NR techniques, to reduce the noise efficiently, while preserving the speech spectral components. The Spectral Subtraction (SS) method that employs the two-channel Generalized Sidelobe Canceller (2chGSC+SS) is suggested to be employed by the proposed technique for processing the signal in the low frequency region where the dominant cues of the speech signals locate. For the high frequency region, the two-sensor NR technique based on the modified Cross-Spectral Subtraction (2chCSS) is employed. Simulation results in the one-noise-source and multiple-noise-source environments are undertaken, while objective and subjective performance measures are used to confirm the potential of the proposed method. In addition, two approaches are proposed in order to control the noise subtraction parameter to be variable and frequency-dependent so that the amount of noise reduction and speech distortion can be managed for the SS-based NR methods. Simulation results are demonstrated where the SS method that employs these proposed variable parameters can sufficiently reduce the additive background noise while the ability to preserve speech quality is superior to the other investigated NR techniques.