Highly-Reverberant Real Environment database: HRRE

Speech recognition in highly-reverberant real environments remains a major challenge. An evaluation dataset for this task is needed. This report describes the generation of the Highly-Reverberant Real Environment database (HRRE). This database contains 13.4 hours of data recorded in real reverberant environments and consists of 20 different testing conditions which consider a wide range of reverberation times and speaker-to-microphone distances. These evaluation sets were generated by re-recording the clean test set of the Aurora-4 database which corresponds to five loudspeaker-microphone distances in four reverberant conditions. Database Recording To generate the data for the test set, we re-recorded the original clean test data from the Aurora-4 database (i.e. 330 utterances recorded with the Sennheiser microphone) in a reverberation chamber considering different speakermicrophone distances and reverberation times (RTs) and following the procedures specified by the ISO 354:2003 Standard [1]. The reverberation chamber has an internal surface area of 100 m, a volume of 63 m and an RTmid equal to three seconds. Four reverberant conditions were generated by adding sound-absorbing materials in the reflecting surfaces of the chamber. The