Alex Graves大神的论文

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