APPARATUS AND METHOD FOR CODING OF VOICE SIGNAL USING NON NEGATIVE FACTORIZATION ALGORITHM

摘要

PURPOSE: A device and a method for coding a voice signal using a non negative factorization (NMF) algorithm are provided to be improved in performance by using the correlation of LSF data which is in the same dimension as a previous frame and the intra-correlation of surrounding frames for voice signal coding. CONSTITUTION: A frequency converter (110) converts each frame of a voice signal into a linear spectrum frequency. An NMF prediction unit (120) produces the prediction frame of a current frame using an NMF prediction algorithm with respect to the previous frame of the voice signal. A vector quantization unit (130) divides a removal frame removed with the prediction frame from the current frame into a plurality of dimensions, performs vector quantization, and reconstructs the current frame through the recombination in the dimension of the removal frame and the addition of the prediction frame. [Reference numerals] (110) Frequency converter; (120) NMF prediction unit; (131) Invalidation unit; (132) Decryption unit; (AA) Input an audio signal; (BB) Output a restored audio signal