Diffusion-Based Voice Conversion with Fast Maximum Likelihood Sampling Scheme

TL;DR

This paper introduces a diffusion probabilistic model for high-quality one-shot voice conversion, enhancing speed with a novel SDE solver suitable for real-time applications, supported by empirical and theoretical analysis.

Contribution

It proposes a scalable diffusion-based voice conversion method and a new SDE solver to accelerate diffusion models without sacrificing quality.

Findings

Superior voice conversion quality compared to state-of-the-art methods

The novel SDE solver improves sampling speed for diffusion models

Theoretical analysis supports the effectiveness of the proposed solver

Abstract

Voice conversion is a common speech synthesis task which can be solved in different ways depending on a particular real-world scenario. The most challenging one often referred to as one-shot many-to-many voice conversion consists in copying the target voice from only one reference utterance in the most general case when both source and target speakers do not belong to the training dataset. We present a scalable high-quality solution based on diffusion probabilistic modeling and demonstrate its superior quality compared to state-of-the-art one-shot voice conversion approaches. Moreover, focusing on real-time applications, we investigate general principles which can make diffusion models faster while keeping synthesis quality at a high level. As a result, we develop a novel Stochastic Differential Equations solver suitable for various diffusion model types and generative tasks as shown through empirical studies and justify it by theoretical analysis.