Toward a representative set of "real-life" noises

The focus of this study was to create a database containing a large number of background noises. These noises were analysed on the basis of their spectral and temporal behaviour. The most important dimensions have been determined by using a factor analysis, which reduced the total number of dimensions from 14 (7 octave bands in the spectral domain and 7 in the temporal domain) to 4. One dimension provides information about the amount of amplitude modulation present in the signal. The other dimensions provide spectral information. Based on the output of the factor analysis, a cluster analysis was carried out to reduce the total number of signals (144) to 15 anchor points (i.e., the noise forming the median of the cluster). A selection of these anchor points can be used for fitting of so-called comfort programs in multi-program hearing aids and for the evaluation of signal-processing schemes in specific background noise conditions.