O NT HEAPPLICATION OF PSYCHOACOUSTICALLY-MOTIVATED DEREVERBERATION FOR RECORDINGS TAKEN IN THE GERMAN PARLIAMENT

In this paper, we discuss the application of speech dereverberation tech- niques for post-processing of recordings taken in the German parliament. Based on a novel psychoacoustically-motivated dereverberation concept, a significant improve- ment in terms of the perceived quality is obtained in comparison to a conventional dereverberation approach. Since time-varying changes of the acoustical environment are negligible, all required acoustical parameters such as reverberation time (RT) and direct-to-reverberant-energy ratio (DRR), are determined in an off-line procedure. In this contribution, we show how such techniques can be used to enhance speech recordings taken in large conference halls or similar environments. For a case study, speech recordings from the German Bundestag, which is located in the Reichstag in Berlin, are investigated and suitable signal processing methods for a post-processing are proposed. Based on a conven- tional dereverberation algorithms, an improved concept which exploits masking properties of the human auditory system is developed. In the remainder of this paper, the acoustical environment is analyzed in the next section which includes the considered signal model, followed by a short discussion on the reverberation time (RT) and direct-to-reverberant energy ratio (DRR). In Section 3, a conventional dereverberation algorithm is introduced and a psychoacoustically-motivated modification is discussion. Finally, in Sections 4 and 5 we show simulation results and draw conclusions. 2 Analysis of the Acoustical Environment Room reverberation is usually caused by reflections of the emitted source, e.g., a speaker which stands far away from another speaker in an enclosure. In contrast to that, in a parliament dis- cussion, the speaker is located at a lectern and the speech is captured by microphones at a small distance. This speech signal is then processed and emitted by a loudspeaker system to the audience. In very large rooms, this signal is then reflected on the walls and fed back into the mi- crophones with a certain sound propagation delay. In order to avoid instability and overshoots, a feedback cancellation is employed which usually consists of a notch fi lter or an adaptivefilter