LANGUAGE-SEPARATION, LANGUAGE-SEPARATION PROCEDURES, LANGUAGE-SEPARATION PROGRAMME AND LANGUAGE-SEPARATION SYSTEM

摘要

A spoken-language separation device (12) of a spoken-language separation system includes a characteristic quantity extraction unit (121),set up to extract time series data of a spoken-language characteristic quantity of mixed-spoken language, a block-sharing unit (122),set up to share the time series data of the spoken-language characteristic quantity in blocks, having a certain time width, a spoken-language separation-neuronal network (1b),set up to create time series data of a mask from each of a variety of speakers from the time series data of the volume of spoken-language characteristics divided into blocks, a spoken-language restoration unit (123),set up to recover the spoken-language data of the spoken language from each of the multitude of speakers from the time series data of the mask and the time series data from the spoken-language characteristics of the mixed language. The Spoken Language Separation Neuronal Network (1b) is established, when creating the time series data of the mask from each of the multitude of speakers, the time series data of a block temporally earlier than a current time in the forward LSTM neuronal network, and the time series data of a block, consists of a predetermined number of frames,time later than the current time in the reverse LSTM neuronal network.