Speech Databases

Along my research, I was recording reverberant speech databases. Here it is published online for the use of the speech enhancement community. The two databases were recorded in the BIU Acoustic Lab with variable reverberation level, using loudspeakers (RevStat) or real humans (RevDyn) as the source for speech signal.

The databases are free for download

RevStat

diagram_RevStat

  •  Eight-microphone array
  • Speech source - loudspeakers positioned in 6 different locations
  • Three different reverberation times - 480ms, 630ms, and 940ms
  • Eight different clean speech signals
  • Total of 6 X 3 X 8 = 144 speech recordings (1 min. each)
  • Noise signals - AC and babble - were also recorded

RevDyn

diagram_RevDyn

  • Human speakers moving and talking in reverberant room
  • Eight-microphone array
  • Speech source - four different English speakers
  • Reverberation time - 750ms
  • Twelve different dynamic scenarios
  • Total of 4 X 12 = 48 speech recordings (1 min. each)