SOUND SOURCE EMPHASIS LEARNING DEVICE, SOUND SOURCE EMPHASIS DEVICE, SOUND SOURCE EMPHASIS LEARNING METHOD, PROGRAM, AND SIGNAL PROCESSING LEARNING DEVICE

Title:

SOUND SOURCE EMPHASIS LEARNING DEVICE, SOUND SOURCE EMPHASIS DEVICE, SOUND SOURCE EMPHASIS LEARNING METHOD, PROGRAM, AND SIGNAL PROCESSING LEARNING DEVICE

Document Type and Number:

Japanese Patent JP2018031910

Kind Code:

A

Abstract:

PROBLEM TO BE SOLVED: To provide a sound source emphasis learning device configured to learn a network parameter for sound source emphasis in an accurate and stable manner.SOLUTION: A previous learning section includes: a target sound non-negative auto-encoder calculation part 126 for calculating a target sound spectrum characteristic vector by modeling a spectral characteristic of a target sound from target sound BF output power that is generated from frequency domain mixed sound learning data; a noise non-negative auto-encoder calculation part 127 for calculating a noise spectral characteristic vector by modeling a spectral characteristic of noise from noise BF output power that is generated from the frequency domain mixed sound learning data; a complementary subtraction neural network calculation part 128 for calculating an estimated target sound PSD and an estimated noise PSD from the target sound spectral characteristic vector and the noise spectral characteristic vector; and a network parameter optimization part 129 for optimizing a network parameter by using the estimated target sound PSD, the estimated noise PSD, a target sound PSD and a noise PSD.SELECTED DRAWING: Figure 2

More Like This:

WO/2017/112343	AUDIO SIGNAL PROCESSING IN NOISY ENVIRONMENTS
WO/2023/052345	AUDIO SOURCE SEPARATION
JP3364487	PROBLEM TO BE SOLVED: To provide a voice separation method for composite voice data, a speaker identification method, a voice separation device for composite voice data, a speaker identification device, a computer program, and a recording medium.

Inventors:

NIWA KENTA
KOBAYASHI KAZUNORI
KOIZUMI YUMA
KAWASE TOMOKO

Application Number:

JP2016164726A

Publication Date:

March 01, 2018

Filing Date:

August 25, 2016

Export Citation:

Click for automatic bibliography generation Help

Assignee:

NIPPON TELEGRAPH & TELEPHONE

International Classes:

G10L21/0272; G10L21/0308

Domestic Patent References:

JP2016143042A	2016-08-08
JP2010541010A	2010-12-24

Foreign References:

US20160071526A1	2016-03-10
US20150095026A1	2015-04-02

Other References:

丹羽健太他: "相互情報量増大型受音とＤＮＮ回帰に基づく遠距離収音技術に関する検討", 日本音響学会２０１６年春季研究発表会講演論文集［ＣＤ−ＲＯＭ］, JPN6019028738, February 2016 (2016-02-01), pages 551 - 554, ISSN: 0004083097

Attorney, Agent or Firm:

Naoki Nakao
Yukio Nakamura
Yoshimura Munehiro

Previous Patent: A speech processing unit, a voice processing method, and a speech processing program

Next Patent: An image forming device and an image forming system