Adaptive Gain Methods to Improve Speech Intelligibility under Reverberation

The reflection of an acoustic signal on walls or objects in an enclosed environment is perceived as reverbera- tion. It is present daily in conference rooms, tunnels and any closed spaces. The presence of reverberation can degrade speech intelligibility and the performance of tasks that depend on it, such as speech or speaker recognition. Many methods were proposed to attenuate this degradation. A family of such methods acts on clean speech, applying gains on parts of the signal in order to improve intelligible when reverberated. These methods are called adaptive gain methods. This study aims to evaluate the effect of two adaptive gain methods, Adaptive Gain Control (AGC) and Steady-State Suppression (SSS), in speech intelligibility. The evaluation uses four objective measures: Coherence Signal Intelligibility Index (CSII), Short Time Objective Intelligibility (STOI), Speech Reverberation to Modulation Ratio (SRMR) and Weighted Short Time Objective Intelligibility (WSTOI). Results show that AGC improves speech intelligibility in studied conditions, while SSS degrades it, a result in line with subjective measures found in the literature.