ACOUSTIC EVENT IDENTIFICATION MODEL LEARNING DEVICE, ACOUSTIC EVENT DETECTION DEVICE, ACOUSTIC EVENT IDENTIFICATION MODEL LEARNING METHOD, ACOUSTIC EVENT DETECTION METHOD, AND PROGRAM

摘要

PROBLEM TO BE SOLVED: To improve accuracy of acoustic event detection.SOLUTION: A learning data featured value extraction part 15 extracts a learning data featured value from learning data as an acoustic signal in which a plurality of acoustic events are included. A model parameter learning part 17 learns a parameter of each identification layer of a deep neutral network for inputting an acoustic signal, and for outputting the likelihood of each acoustic event by using the learning data featured value. A development data featured value extraction part 16 extracts a development data featured value from development data as the acoustic signal in which the plurality of acoustic events are included. A model parameter adjustment part 18 optimizes the parameter of each layer of the deep neural network by using the learning data featured value, the development data featured value, a correct label of the learning data featured value and a correct label of the development data featured value.