Thispaperpresentsaspeechrecognitionsys-temthatdirectlytranscribesaudiodatawithtext,withoutrequiringanintermediatephoneticrepre-sentation.ThesystemisbasedonacombinationofthedeepbidirectionalLSTMrecurrentneuralnetworkarchitectureandtheConnectionistTem-poralClassificationobjectivefunction.