This paper describes the speech recognition systems of IOITfor IWSLT 2014 TED ASR track. This year, we focus onimproving acoustic model for the systems using two mainapproaches of deep neural network which are hybrid and bottleneck feature systems. These two subsystems are combinedusing lattice Minimum Bayes-Risk decoding. On the 2013evaluations set, which serves as a progress test set, we wereable to reduce the word error rate of our transcription systemsfrom 27.2% to 24.0%, a relative reduction of 11.7%.