Hearing apparatuses, in particular hearing aids, have improved intelligibility. This is accomplished by a method for amplifying a speech signal for a hearing apparatus by detection of an input sound, determining an input level of the input sound and amplification of the input sound. A mean input level of the input sound is determined and it is established whether the input sound has a speech component. If not, the input signal is amplified in dependence on the input level in accordance with a predefined first amplification characteristic. However, if a speech component is found, the input signal is amplified in dependence on the input level in accordance with a predefined second amplification characteristic. The second amplification characteristic is formed in dependence on the mean input level such that amplification is increased with respect to first amplification characteristic in a predefined level interval below the mean input level.