Parallel-Data-Free Voice Conversion Using Cycle-Consistent Adversarial Networks

TL;DR

This paper introduces CycleGAN-VC, a novel voice conversion method that learns mappings from unpaired source and target speech data without needing parallel datasets, avoiding over-smoothing and maintaining high quality.

Contribution

It presents a parallel-data-free voice conversion approach using cycle-consistent adversarial networks with gated CNNs, improving quality and avoiding over-smoothing compared to traditional methods.

Findings

Converted speech quality is comparable to parallel-data-based methods.

Objective metrics indicate near-natural feature sequences.

Method effectively learns mappings from unpaired data.

Abstract

We propose a parallel-data-free voice-conversion (VC) method that can learn a mapping from source to target speech without relying on parallel data. The proposed method is general purpose, high quality, and parallel-data free and works without any extra data, modules, or alignment procedure. It also avoids over-smoothing, which occurs in many conventional statistical model-based VC methods. Our method, called CycleGAN-VC, uses a cycle-consistent adversarial network (CycleGAN) with gated convolutional neural networks (CNNs) and an identity-mapping loss. A CycleGAN learns forward and inverse mappings simultaneously using adversarial and cycle-consistency losses. This makes it possible to find an optimal pseudo pair from unpaired data. Furthermore, the adversarial loss contributes to reducing over-smoothing of the converted feature sequence. We configure a CycleGAN with gated CNNs and train it with an identity-mapping loss. This allows the mapping function to capture sequential and hierarchical structures while preserving linguistic information. We evaluated our method on a parallel-data-free VC task. An objective evaluation showed that the converted feature sequence was near natural in terms of global variance and modulation spectra. A subjective evaluation showed that the quality of the converted speech was comparable to that obtained with a Gaussian mixture model-based method under advantageous conditions with parallel and twice the amount of data.