Multicondition Training Set --------------------------- Half(3569) of WSJ0(7138) utterances recorded with the Sennheiser microphone and the other half recorded with the second microphone. No noise added to one fourth(893) of each of the two subsets. Artificial noise is added to rest of the data. The noise type is randomly chosen out of 6 noises (Car, Babble, Restaurant, Street, Airport and Train Stations) and at randomly chosen SNR between 10 and 20 dB in steps of 1 dB. Acoustic Models: cross-word with 4 mixtures per state Training Time: 55 Hours on an 800 MHz processor