Title:
SIGNAL PROCESSING DEVICE, SIGNAL PROCESSING METHOD, AND SIGNAL PROCESSING PROGRAM
Document Type and Number:
WIPO Patent Application WO/2022/168251
Kind Code:
A1
Abstract:
A signal processing device (100) comprises: an acquisition unit (120) that acquires an object input signal indicating a mixed sound including a target sound, and a learned model; a feature amount extraction unit (130) that, on the basis of the object input signal, extracts a feature amount sequence indicating a plurality of feature amounts; a feature amount normalization unit (140) that calculates a temporary normalization parameter on the basis of the feature amount sequence, corrects the temporary normalization parameter using a preset correction method, and normalizes the feature amount sequence using a corrected normalization parameter obtained by the correction; a calculation unit (150) that calculates a target sound feature amount sequence indicating a plurality of feature amounts of the target sound using a normalized feature amount sequence obtained by the normalization, and the learned model; and a signal generation unit (160) that generates an object output signal indicating the target sound on the basis of the target sound feature amount sequence.
Inventors:
MITSUI YOSHIKI (JP)
Application Number:
PCT/JP2021/004220
Publication Date:
August 11, 2022
Filing Date:
February 05, 2021
Export Citation:
Assignee:
MITSUBISHI ELECTRIC CORP (JP)
International Classes:
G10L21/0308
Foreign References:
JP2008311866A | 2008-12-25 | |||
US20190318757A1 | 2019-10-17 | |||
US20190066713A1 | 2019-02-28 |
Other References:
LIN, KIN WAH ET AL.: "ZERO-MEAN CONVOLUTIONAL NETWORK WITH DATA AUGMENTATION FOR SOUND LEVEL INVARIANT SINGING VOICE SEPARATION", 2019 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING, 17 April 2019 (2019-04-17), XP033565538
Attorney, Agent or Firm:
YAMAGATA Yoichi et al. (JP)
Download PDF: