AdaptVC: High Quality Voice Conversion with Adaptive Learning

TL;DR

AdaptVC introduces a novel voice conversion method that employs adaptive learning with self-supervised features and a conditional flow matching decoder, achieving high-quality, zero-shot voice conversion with improved robustness and similarity.

Contribution

The paper presents a new approach using adapters and a conditional flow matching decoder for disentangling content and speaker features in voice conversion, enhancing zero-shot performance.

Findings

Outperforms existing models in speech quality and similarity in zero-shot scenarios.

Achieves effective disentanglement of content and speaker features.

Demonstrates robustness and efficiency in voice conversion tasks.

Abstract

The goal of voice conversion is to transform the speech of a source speaker to sound like that of a reference speaker while preserving the original content. A key challenge is to extract disentangled linguistic content from the source and voice style from the reference. While existing approaches leverage various methods to isolate the two, a generalization still requires further attention, especially for robustness in zero-shot scenarios. In this paper, we achieve successful disentanglement of content and speaker features by tuning self-supervised speech features with adapters. The adapters are trained to dynamically encode nuanced features from rich self-supervised features, and the decoder fuses them to produce speech that accurately resembles the reference with minimal loss of content. Moreover, we leverage a conditional flow matching decoder with cross-attention speaker conditioning to further boost the synthesis quality and efficiency. Subjective and objective evaluations in a zero-shot scenario demonstrate that the proposed method outperforms existing models in speech quality and similarity to the reference speech.