This 4th CHiME Speech Separation and Recognition Challenge (CHiME-4) challenge revisits the datasets originally recorded for CHiME-3, i.e., Wall Street Journal corpus sentences spoken by talkers situated in challenging noisy environments recorded using a 6-channel tablet based microphone array. CHiME-4 increases the level of difficulty by constraining the number of microphones available for testing.