Effects of suppressing steady-state portions of speech on intelligibility in reverberant environments

1. Introduction When listening to a lecture in a large auditorium it is often difficult to understand the speech. Among other factors, comprehension may be impaired by reverberation, which is sound reflecting from the wall, interfering with direct sound. Based on the modulation transfer function (MTF), the speech transmission index (STI) has been proposed as an objective measure for speech intelligibility in rooms [1]. There are two types of processing for improving speech intelligibility under reverberant conditions. One is called pre-processing, where signal processing is done before a signal is radiated through a loudspeaker to prevent intelligibility degradation. The other is called post-processing, where signal processing is done on reverberant speech. Langhans et al. [2] proposed a method for post-processing using a single microphone, and Avendano et al. [3] extended their method. Both researchers used so-called ''modulation filtering,'' or filtering on a temporal envelope, after dividing the speech signal into a series of subbands. Langhans et al. [2] also tried modulation filtering for the pre-processing method. In their study, however, a large improvement was not reported. Our group has also tested speech intelligibility based on a similar type of modulation filtering [4–7]. In our previous studies, we emphasized modulation characteristics of speech around 3–6 Hz which corresponds to the syllabic rate of speech, and as a result, we observed some tendencies toward improvement, although they were not significant. In the present study, we focus on suppressing the steady-state portions of speech to investigate the effect of modulation filtering on speech intelligibility. In light of the fact that modulation filtering essentially suppresses steady-state portions and emphasizes speech dynamics, we hypothesize that steady-state suppressed speech sounds are more robust in the case of reverberation. To test this hypothesis, we compare intelligibilities of speech after adding reverberation to speech signals with and without steady-state suppression.