A fundamental quantity to measure is the long-term RMS level of the speech and noise signals. The ratio between these (SNR) is not necessarily the same at the output as it is at the input of the compressor. As an example, Figure 1 shows the relationship between Input SNR and Output SNR for a fast-acting compressor, when the noise is highly modulated. It can be seen that the Output SNR may be either lower or higher than the Input SNR, dependent on the Input SNR itself. The change in long-term SNR through the system may be several dB in either direction.
It is natural to ask whether objective effects of the type seen in Figure 1 are reflected in the speech intelligibility performance of real listeners. To test this, we carried out a listening experiment with twelve hearing-impaired subjects. Seven different conditions were constructed, such that the long-term SNR at the ear (Output SNR) varied systematically as a combination of variations in Input SNR and variations of compression system parameters. We found that a given alteration in Output SNR caused by compression leads to a change in listener performance similar to that which occurs if the Input SNR (in a linear hearing aid) is altered by the same amount.
Further work has investigated the utility of various alternative indices for predicting the intelligibility changes observed in the abovementioned listening tests. So far, the Extended Speech Intelligibility Index (ESII)2 seems to be an excellent candidate.