1. Erasmus University Medical Center, Rotterdam, the Netherlands
2. Academic Medical Center, Amsterdam, the Netherlands

In an earlier talk we stressed the importance of a method to describe the effects of compression on the speech signal. A model to describe the effect of general signal processing on a speech signal is the Speech Transmission Index (STI). The method makes a clear distinction between the spectral information (formants, spectral bands of energy, of which the relevant parts range from 250 Hz to 8000Hz for speech) and the modulation information in these spectral bands (the relevant frequencies range from 0.5 to about 32 Hz). However, the method is not suited to describe nonlinear processing like compression as no information is lost to reverberation or background noise. We used a method similar to the STI analysis to describe the amount of transferred information on modulations. The analysis was split into two parts. First we analyzed the compressor by inputting a modulated sine-wave and comparing the modulation depth at output and input. The method gives a clear indication of the frequency range of the modulations that are suppressed and the modulation frequencies that are still present in the signal. Secondly, we determined the effect of compression on the distribution of the speech levels, thus defined an effective compression ratio for speech. Both methods give an indication of the effectiveness of the compressor in relation to a speech signal. They do not only take into account the static characteristics as given in datasheet, but also the temporal behavior that is often only described in terms of attack and release times for each frequency band. It is rather difficult to interpret a number of time constants on their relevance to speech. This described approach can indicate what information (spectral, temporal) may be affected by a compression system. It will indicate the type of perceptual tests that should be done to test positive or negative effects on speech intelligibility.