Digital Enhancement of Speech Perception in Noisy Environments

The perception of loudspeaker speech from public address systems or mobile phones is often affected by local background noise. In order to improve the speech intelligibility, “Near-End Listening Enhancement” (NELE) can be applied. This processing adaptively modifies the loudspeaker signal by taking the instantaneous background noise into account. The main novel contributions of this thesis are NELE frequency domain weighting rules without need for complex optimization and NELE time domain filter structures with low latency and complexity. Furthermore, NELE solutions for public address systems with massive crosstalk, such as railway stations, are presented. In addition, solutions for combined operation of NELE with far-end noise reduction are developed, which are relevant for mobile telephony. Listening tests as well as objective measures confirm a significant increase in intelligibility.