Speech classification apparatus, speech classification method, program

摘要

PROBLEM TO BE SOLVED: To provide a voice classification device capable of classifying whether a voice is in a reading tone or in a conversation tone with high accuracy.SOLUTION: A voice classification device includes: a sound feature quantity sequence extraction unit for extracting N sound feature quantity sequences in total from each of N-pieces of speech, with N being an integer equal to or greater than 2; a base speech model storage unit for storing a base speech model; a mixed normal distribution adaptation unit for generating N speech models by adapting the base speech model to each of the N sound feature quantity sequences; a super vector calculation unit for outputting N speech super vectors from the N speech models; a dispersion calculation unit for generating a speech fluctuation vector from the N speech super vectors; a model parameter storage unit for storing a model parameter of an identifier; and a tone discrimination unit for generating a tone discrimination result for each of the N-pieces of speech, based on the speech fluctuation vector the model parameter of the identifier.