首页>
外国专利>
ACOUSTIC MODEL LEARNING METHOD, VOICE RECOGNITION METHOD, ACOUSTIC MODEL LEARNING DEVICE, VOICE RECOGNITION DEVICE, ACOUSTIC MODEL LEARNING PROGRAM, AND VOICE RECOGNITION PROGRAM
ACOUSTIC MODEL LEARNING METHOD, VOICE RECOGNITION METHOD, ACOUSTIC MODEL LEARNING DEVICE, VOICE RECOGNITION DEVICE, ACOUSTIC MODEL LEARNING PROGRAM, AND VOICE RECOGNITION PROGRAM
An acoustic model learning device (10) first extracts a voice feature quantity indicating a feature of voice data, and then, on the basis of an acoustic condition calculation model parameter that characterizes the calculation model of an acoustic condition represented by a neural network, calculates an acoustic condition feature quantity indicating a feature of an acoustic condition of the voice data using an acoustic condition calculation model. Next, the acoustic model learning device (10) generates a corrected parameter, which is a parameter obtained by correcting, on the basis of the acoustic condition feature quantity, an acoustic model parameter that characterizes an acoustic model represented by a neural network to which an output layer of the acoustic condition calculation model is joined. The acoustic model learning device (10) then updates the acoustic model parameter on the basis of the corrected parameter and the voice feature quantity, and updates the acoustic condition calculation model parameter on the basis of the corrected parameter and the voice feature quantity.
展开▼
