CycleGAN-Based High-Quality Non-Parallel Voice Conversion with Spectrogram and WaveRNN

Aoi Kanagaki; Masaya Tanaka; Takashi Nose; Ryohei Shimizu; Akira Ito; Akinori Ito

doi:10.1109/GCCE50665.2020.9291952

CycleGAN-Based High-Quality Non-Parallel Voice Conversion with Spectrogram and WaveRNN

Aoi Kanagaki, Masaya Tanaka, Takashi Nose, Ryohei Shimizu, Akira Ito, Akinori Ito

ENG - Communications Engineering

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

1 Citation (Scopus)

Abstract

This paper proposes Scyclone, a high-quality voice conversion (VC) technique without parallel data training. Scyclone improves speech naturalness and speaker similarity of the converted speech by introducing CycleGAN-based spectrogram conversion with a simplified WaveRNN-based vocoder. In Scyclone, a linear spectrogram is used as the conversion feature, which avoids quality degradation due to extraction errors. The subjective experiments show that Scyclone is significantly better than CycleGAN-VC2, one of the existing state-of-the-art parallel-data-free VC techniques.

Original language	English
Title of host publication	2020 IEEE 9th Global Conference on Consumer Electronics, GCCE 2020
Publisher	Institute of Electrical and Electronics Engineers Inc.
Pages	356-357
Number of pages	2
ISBN (Electronic)	9781728198026
DOIs	https://doi.org/10.1109/GCCE50665.2020.9291952
Publication status	Published - 2020 Oct 13
Event	9th IEEE Global Conference on Consumer Electronics, GCCE 2020 - Kobe, Japan Duration: 2020 Oct 13 → 2020 Oct 16

Publication series

Name	2020 IEEE 9th Global Conference on Consumer Electronics, GCCE 2020

Conference

Conference	9th IEEE Global Conference on Consumer Electronics, GCCE 2020
Country/Territory	Japan
City	Kobe
Period	20/10/13 → 20/10/16

Keywords

CycleGAN
Voice conversion (VC)
WaveRNN
parallel-data-free VC
spectrogram

ASJC Scopus subject areas

Signal Processing
Electrical and Electronic Engineering
Media Technology
Instrumentation
Computer Networks and Communications
Computer Vision and Pattern Recognition

Access to Document

10.1109/GCCE50665.2020.9291952

Cite this

Kanagaki, A., Tanaka, M., Nose, T., Shimizu, R., Ito, A., & Ito, A. (2020). CycleGAN-Based High-Quality Non-Parallel Voice Conversion with Spectrogram and WaveRNN. In 2020 IEEE 9th Global Conference on Consumer Electronics, GCCE 2020 (pp. 356-357). [9291952] (2020 IEEE 9th Global Conference on Consumer Electronics, GCCE 2020). Institute of Electrical and Electronics Engineers Inc.. https://doi.org/10.1109/GCCE50665.2020.9291952

CycleGAN-Based High-Quality Non-Parallel Voice Conversion with Spectrogram and WaveRNN. / Kanagaki, Aoi; Tanaka, Masaya; Nose, Takashi et al.
2020 IEEE 9th Global Conference on Consumer Electronics, GCCE 2020. Institute of Electrical and Electronics Engineers Inc., 2020. p. 356-357 9291952 (2020 IEEE 9th Global Conference on Consumer Electronics, GCCE 2020).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Kanagaki, A, Tanaka, M, Nose, T, Shimizu, R, Ito, A & Ito, A 2020, CycleGAN-Based High-Quality Non-Parallel Voice Conversion with Spectrogram and WaveRNN. in 2020 IEEE 9th Global Conference on Consumer Electronics, GCCE 2020., 9291952, 2020 IEEE 9th Global Conference on Consumer Electronics, GCCE 2020, Institute of Electrical and Electronics Engineers Inc., pp. 356-357, 9th IEEE Global Conference on Consumer Electronics, GCCE 2020, Kobe, Japan, 20/10/13. https://doi.org/10.1109/GCCE50665.2020.9291952

Kanagaki A, Tanaka M, Nose T, Shimizu R, Ito A, Ito A. CycleGAN-Based High-Quality Non-Parallel Voice Conversion with Spectrogram and WaveRNN. In 2020 IEEE 9th Global Conference on Consumer Electronics, GCCE 2020. Institute of Electrical and Electronics Engineers Inc. 2020. p. 356-357. 9291952. (2020 IEEE 9th Global Conference on Consumer Electronics, GCCE 2020). doi: 10.1109/GCCE50665.2020.9291952

@inproceedings{e93c74df35884c8a8728a03de02f8351,

title = "CycleGAN-Based High-Quality Non-Parallel Voice Conversion with Spectrogram and WaveRNN",

abstract = "This paper proposes Scyclone, a high-quality voice conversion (VC) technique without parallel data training. Scyclone improves speech naturalness and speaker similarity of the converted speech by introducing CycleGAN-based spectrogram conversion with a simplified WaveRNN-based vocoder. In Scyclone, a linear spectrogram is used as the conversion feature, which avoids quality degradation due to extraction errors. The subjective experiments show that Scyclone is significantly better than CycleGAN-VC2, one of the existing state-of-the-art parallel-data-free VC techniques.",

keywords = "CycleGAN, Voice conversion (VC), WaveRNN, parallel-data-free VC, spectrogram",

author = "Aoi Kanagaki and Masaya Tanaka and Takashi Nose and Ryohei Shimizu and Akira Ito and Akinori Ito",

note = "Funding Information: A part of this work was supported by the JST COI Grant Number JPMJCE1303 and the JSPS Grant-in-Aid for Scientific Research JP17H00823. Publisher Copyright: {\textcopyright} 2020 IEEE.; 9th IEEE Global Conference on Consumer Electronics, GCCE 2020 ; Conference date: 13-10-2020 Through 16-10-2020",

year = "2020",

month = oct,

day = "13",

doi = "10.1109/GCCE50665.2020.9291952",

language = "English",

series = "2020 IEEE 9th Global Conference on Consumer Electronics, GCCE 2020",

publisher = "Institute of Electrical and Electronics Engineers Inc.",

pages = "356--357",

booktitle = "2020 IEEE 9th Global Conference on Consumer Electronics, GCCE 2020",

}

TY - GEN

T1 - CycleGAN-Based High-Quality Non-Parallel Voice Conversion with Spectrogram and WaveRNN

AU - Kanagaki, Aoi

AU - Tanaka, Masaya

AU - Nose, Takashi

AU - Shimizu, Ryohei

AU - Ito, Akira

AU - Ito, Akinori

PY - 2020/10/13

Y1 - 2020/10/13

N2 - This paper proposes Scyclone, a high-quality voice conversion (VC) technique without parallel data training. Scyclone improves speech naturalness and speaker similarity of the converted speech by introducing CycleGAN-based spectrogram conversion with a simplified WaveRNN-based vocoder. In Scyclone, a linear spectrogram is used as the conversion feature, which avoids quality degradation due to extraction errors. The subjective experiments show that Scyclone is significantly better than CycleGAN-VC2, one of the existing state-of-the-art parallel-data-free VC techniques.

AB - This paper proposes Scyclone, a high-quality voice conversion (VC) technique without parallel data training. Scyclone improves speech naturalness and speaker similarity of the converted speech by introducing CycleGAN-based spectrogram conversion with a simplified WaveRNN-based vocoder. In Scyclone, a linear spectrogram is used as the conversion feature, which avoids quality degradation due to extraction errors. The subjective experiments show that Scyclone is significantly better than CycleGAN-VC2, one of the existing state-of-the-art parallel-data-free VC techniques.

KW - CycleGAN

KW - Voice conversion (VC)

KW - WaveRNN

KW - parallel-data-free VC

KW - spectrogram

UR - http://www.scopus.com/inward/record.url?scp=85099348438&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85099348438&partnerID=8YFLogxK

U2 - 10.1109/GCCE50665.2020.9291952

DO - 10.1109/GCCE50665.2020.9291952

M3 - Conference contribution

AN - SCOPUS:85099348438

T3 - 2020 IEEE 9th Global Conference on Consumer Electronics, GCCE 2020

SP - 356

EP - 357

BT - 2020 IEEE 9th Global Conference on Consumer Electronics, GCCE 2020

PB - Institute of Electrical and Electronics Engineers Inc.

T2 - 9th IEEE Global Conference on Consumer Electronics, GCCE 2020

Y2 - 13 October 2020 through 16 October 2020

ER -

CycleGAN-Based High-Quality Non-Parallel Voice Conversion with Spectrogram and WaveRNN

Abstract

Publication series

Conference

Keywords

ASJC Scopus subject areas

Access to Document

Other files and links

Fingerprint

Cite this