LatentVoiceGrad: Nonparallel Voice Conversion with Latent Diffusion/Flow-Matching Models

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Kameoka, Hirokazu, Kaneko, Takuhiro, Tanaka, Kou, Kondo, Yuto
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866912580500455424
author Kameoka, Hirokazu
Kaneko, Takuhiro
Tanaka, Kou
Kondo, Yuto
author_facet Kameoka, Hirokazu
Kaneko, Takuhiro
Tanaka, Kou
Kondo, Yuto
contents Previously, we introduced VoiceGrad, a nonparallel voice conversion (VC) technique enabling mel-spectrogram conversion from source to target speakers using a score-based diffusion model. The concept involves training a score network to predict the gradient of the log density of mel-spectrograms from various speakers. VC is executed by iteratively adjusting an input mel-spectrogram until resembling the target speaker's. However, challenges persist: audio quality needs improvement, and conversion is slower compared to modern VC methods designed to operate at very high speeds. To address these, we introduce latent diffusion models into VoiceGrad, proposing an improved version with reverse diffusion in the autoencoder bottleneck. Additionally, we propose using a flow matching model as an alternative to the diffusion model to further speed up the conversion process without compromising the conversion quality. Experimental results show enhanced speech quality and accelerated conversion compared to the original.
format Preprint
id arxiv_https___arxiv_org_abs_2509_08379
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle LatentVoiceGrad: Nonparallel Voice Conversion with Latent Diffusion/Flow-Matching Models
Kameoka, Hirokazu
Kaneko, Takuhiro
Tanaka, Kou
Kondo, Yuto
Sound
Audio and Speech Processing
Previously, we introduced VoiceGrad, a nonparallel voice conversion (VC) technique enabling mel-spectrogram conversion from source to target speakers using a score-based diffusion model. The concept involves training a score network to predict the gradient of the log density of mel-spectrograms from various speakers. VC is executed by iteratively adjusting an input mel-spectrogram until resembling the target speaker's. However, challenges persist: audio quality needs improvement, and conversion is slower compared to modern VC methods designed to operate at very high speeds. To address these, we introduce latent diffusion models into VoiceGrad, proposing an improved version with reverse diffusion in the autoencoder bottleneck. Additionally, we propose using a flow matching model as an alternative to the diffusion model to further speed up the conversion process without compromising the conversion quality. Experimental results show enhanced speech quality and accelerated conversion compared to the original.
title LatentVoiceGrad: Nonparallel Voice Conversion with Latent Diffusion/Flow-Matching Models
topic Sound
Audio and Speech Processing
url https://arxiv.org/abs/2509.08379