About 10-20 percent of the population suffers from hearing impairment, mainly caused by damage of the inner ear hair cells in the process of aging or by exposure to loud noise. Digital hearing aids compensate individual hearing losses by frequency dependent amplification and compression. However, hearing aid customers suffer severely from reduced speech intelligibility in noisy environments. In this thesis, new digital signal processing algorithms for enhancing acoustically disturbed speech are presented. Besides for hearing aids, the proposed algorithms can also be used for speech communication applications like hands-free telephony or speech recognition.
For systems with one or multiple uncoupled mircophones a frequency domain filter is introduced. The filter achieves an improved quality compared to common systems by applying a statistical speech estimator with an accurate statistical model and is also computationally more efficient.
For systems with coupled microphones the frequency domain statistical speech estimation is generalized resulting in an improved quality especially in difficult acoustic environments with multiple speech interferers.
Finally, for systems with two binaurally coupled microphones, binaural directional filter, which exploit the spatial diversity of speech and noise sources, are introduced. The binaural algorithms can significantly improve the speech intelligibility in a noisy environment.