Title:
VOICE CONVERSION DEVICE, VOICE CONVERSION METHOD, AND VOICE CONVERSION PROGRAM
Document Type and Number:
WIPO Patent Application WO/2021/033685
Kind Code:
A1
Abstract:
Provided is a voice conversion device that uses spectral differentials and can achieve both high sound quality and real-time conversion. The voice conversion device 10 comprises: an acquisition unit 11 that acquires a signal of a voice of a subject; a filter calculation unit 12 that converts feature values representing the vocal tone of the voice using a trained conversion model, multiplies the converted feature values by a trained lifter, and calculates a filter spectrum; a truncated filter calculation unit 13 that performs an inverse Fourier transform on the filter spectrum and applies a prescribed window function to calculate a truncated filter; and a generation unit 14 that multiplies the spectrum of the signal by a spectrum obtained by performing a Fourier transform on the truncated filter and performs an inverse Fourier transform to generate a synthesized voice.
Inventors:
TAKAMICHI SHINNOSUKE (JP)
SAITO YUKI (JP)
SAEKI TAKAAKI (JP)
SARUWATARI HIROSHI (JP)
SAITO YUKI (JP)
SAEKI TAKAAKI (JP)
SARUWATARI HIROSHI (JP)
Application Number:
PCT/JP2020/031122
Publication Date:
February 25, 2021
Filing Date:
August 18, 2020
Export Citation:
Assignee:
UNIV TOKYO (JP)
International Classes:
G10L21/007; G10L25/30
Foreign References:
| JP2018005048A | 2018-01-11 |
Other References:
SUDA, HITOSHI ET AL.: "A Revisit to Feature Handling for High-quality Voice Conversion Based on Gaussian Mixture Model", 2018 ASIA-PACIFIC SIGNAL AND INFORMATION PROCESSING ASSOCIATION ANNUAL SUMMIT AND CONFERENCE, November 2018 (2018-11-01), pages 816 - 822, XP033526014, DOI: 10.23919/APSIPA.2018.8659658
AIKAWA, KIYOAKI ET AL.: "Optimized dynamic cepstrum based on minimum error classification", LECTURE PROCEEDINGS OF ACOUSTICAL SOCIETY OF JAPAN, March 1994 (1994-03-01), pages 123 - 124
SAEKI, TAKAAKI ET AL.: "Lifter learning and subband processing for DNN voice conversion based on a difference-spectrum method", IPSJ SIG TECHNICAL REPORTS, vol. 2020 -SL, no. 2, February 2020 (2020-02-01), pages 1 - 6
AIKAWA, KIYOAKI ET AL.: "Optimized dynamic cepstrum based on minimum error classification", LECTURE PROCEEDINGS OF ACOUSTICAL SOCIETY OF JAPAN, March 1994 (1994-03-01), pages 123 - 124
SAEKI, TAKAAKI ET AL.: "Lifter learning and subband processing for DNN voice conversion based on a difference-spectrum method", IPSJ SIG TECHNICAL REPORTS, vol. 2020 -SL, no. 2, February 2020 (2020-02-01), pages 1 - 6
Attorney, Agent or Firm:
INABA, Yoshiyuki et al. (JP)
Download PDF: