Harmonic-Net: Fundamental frequency and speech rate controllable fast neural vocoder

---

K. Matsubara, T. Okamoto, R. Takashima, T. Takiguchi, T. Toda, and H. Kawai, "Harmonic-Net: Fundamental frequency and speech-rate controllable fast neural vocoder," IEEE/ACM Trans. Audio, Speech, Lang. Process. (accepted, in press)

---

Unseen speaker synthesis with multi-speaker model trained using JVS corpus

Normal condifion

Male (jvs001)

Original	WORLD	uSFGAN	WaveNet

HiFi-GAN	HiFi-GAN (melspc)	Harmonic-Net	Harmonic-Net+

Female (jvs004)

Original	WORLD	uSFGAN	WaveNet

HiFi-GAN	HiFi-GAN (melspc)	Harmonic-Net	Harmonic-Net+

0.5 x fo condifion

Male (jvs001)

WORLD	uSFGAN	HiFi-GAN

Harmonic-Net	Harmonic-Net+

Female (jvs004)

WORLD	uSFGAN	HiFi-GAN

Harmonic-Net	Harmonic-Net+

1.5 x fo condifion

Male (jvs001)

WORLD	uSFGAN	HiFi-GAN

Harmonic-Net	Harmonic-Net+

Female (jvs004)

WORLD	uSFGAN	HiFi-GAN

Harmonic-Net	Harmonic-Net+

0.8 x T condifion

Male (jvs001)

WORLD	WaveNet	HiFi-GAN	HiFi-GAN (melspc)

Harmonic-Net	Harmonic-Net+

Female (jvs004)

WORLD	WaveNet	HiFi-GAN	HiFi-GAN (melspc)

Harmonic-Net	Harmonic-Net+

1.5 x T condifion

Male (jvs001)

WORLD	WaveNet	HiFi-GAN	HiFi-GAN (melspc)

Harmonic-Net	Harmonic-Net+

Female (jvs004)

WORLD	WaveNet	HiFi-GAN	HiFi-GAN (melspc)

Harmonic-Net	Harmonic-Net+

---

Full-band singing voice synthesis using Tohoku Kiritan corpus

Normal condifion

Original	WORLD	HiFi-GAN	HiFI-GAN (melspc)

PeriodNet	Harmonic-Net	Harmonic-Net+

0.5 x fo condifion

WORLD	HiFi-GAN	PeriodNet	Harmonic-Net

Harmonic-Net+

1.5 x fo condifion

WORLD	HiFi-GAN	PeriodNet	Harmonic-Net

Harmonic-Net+

0.8 x T condifion

WORLD	HiFi-GAN	HiFi-GAN (melspc)	PeriodNet

Harmonic-Net	Harmonic-Net+

1.5 x T condifion

WORLD	HiFi-GAN	HiFi-GAN (melspc)	PeriodNet

Harmonic-Net	Harmonic-Net+

---

Text-to-speech using JSUT corpus

Normal condifion

Original	HiFi-GAN	HiFi-GAN (melspc)	Harmonic-Net+

0.5 x fo condifion

HiFi-GAN	Harmonic-Net+

1.5 x fo condifion

HiFi-GAN	Harmonic-Net+