Vocoder-Projected Feature Discriminator

Fuente: arXiv
Guardado en:
Detalles Bibliográficos
Autores principales: Kaneko, Takuhiro, Kameoka, Hirokazu, Tanaka, Kou, Kondo, Yuto
Formato: Preprint
Publicado: 2025
Materias:
Acceso en línea:
Etiquetas: Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
_version_ 1866915465365815296
author Kaneko, Takuhiro
Kameoka, Hirokazu
Tanaka, Kou
Kondo, Yuto
author_facet Kaneko, Takuhiro
Kameoka, Hirokazu
Tanaka, Kou
Kondo, Yuto
contents In text-to-speech (TTS) and voice conversion (VC), acoustic features, such as mel spectrograms, are typically used as synthesis or conversion targets owing to their compactness and ease of learning. However, because the ultimate goal is to generate high-quality waveforms, employing a vocoder to convert these features into waveforms and applying adversarial training in the time domain is reasonable. Nevertheless, upsampling the waveform introduces significant time and memory overheads. To address this issue, we propose a vocoder-projected feature discriminator (VPFD), which uses vocoder features for adversarial training. Experiments on diffusion-based VC distillation demonstrated that a pretrained and frozen vocoder feature extractor with a single upsampling step is necessary and sufficient to achieve a VC performance comparable to that of waveform discriminators while reducing the training time and memory consumption by 9.6 and 11.4 times, respectively.
format Preprint
id arxiv_https___arxiv_org_abs_2508_17874
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Vocoder-Projected Feature Discriminator
Kaneko, Takuhiro
Kameoka, Hirokazu
Tanaka, Kou
Kondo, Yuto
Sound
Artificial Intelligence
Machine Learning
Audio and Speech Processing
In text-to-speech (TTS) and voice conversion (VC), acoustic features, such as mel spectrograms, are typically used as synthesis or conversion targets owing to their compactness and ease of learning. However, because the ultimate goal is to generate high-quality waveforms, employing a vocoder to convert these features into waveforms and applying adversarial training in the time domain is reasonable. Nevertheless, upsampling the waveform introduces significant time and memory overheads. To address this issue, we propose a vocoder-projected feature discriminator (VPFD), which uses vocoder features for adversarial training. Experiments on diffusion-based VC distillation demonstrated that a pretrained and frozen vocoder feature extractor with a single upsampling step is necessary and sufficient to achieve a VC performance comparable to that of waveform discriminators while reducing the training time and memory consumption by 9.6 and 11.4 times, respectively.
title Vocoder-Projected Feature Discriminator
topic Sound
Artificial Intelligence
Machine Learning
Audio and Speech Processing
url https://arxiv.org/abs/2508.17874