Mitigating the effects of reverberation for effective human-robot interaction in the real world

In this paper, we present a method in suppressing speech degradation affecting human-robot communication in real reverberant environment condition. The novelty of the proposed method is the mechanism to indirectly incorporate language information in the enhancement process. This is achieved by considering the effects of the acoustic energy transfer of two consecutive words. The proposed method is categorized into two namely, a one-time computation of the smearing coefficients (offline mode) and the actual enhancement (online mode). In the offline mode, word pair smearing coefficients reflective of the inter-word energy transfer within a word pair are calculated and stored, creating a pool of smearing priors database. Then, the online robust enhancement process integrates this information to the conventional framewise enhancement method. In theory, the proposed method outperforms the conventional framewise-only enhancement since it is able to dynamically update the enhancement parameters based on the actual acoustic energy transfer between successive words during testing. Experiments using a humanoid robot inside a reverberant room confirms the effectiveness of the proposed method. Our method renders human-robot speech communication robust to the effect of reverberation as opposed to the conventional method.